To Switch or Not to Switch?
A new approach to Switching costs
MASTER THESIS
MSc Marketing
Intelligence
To Switch or Not to Switch?
A new approach to Switching costs
Faculty of Economics & Business Department of Marketing Master thesis 18 June 2018 Winschoterdiep 46, 9723 AC Groningen +31685426487 h.venkatraman@student.rug.nl S3464199 Supervisor Dr. A. Bhattacharya Second supervisor Dr. J. (Jenny) van Doorn
ACKNOWLEDGEMENT
I would first like to thank the University of Groningen, specifically the Marketing department for providing me with this fantastic opportunity to study here in Groningen. The basic foundation for carrying out my master thesis was laid by the wonderful faculty of the department and it made my job of writing a thesis that much easier.
I would also like to thank my supervisor Dr. Abhi Bhattacharya for being a fantastic mentor, guiding me through every obstacle that I faced in my thesis. I initially viewed my thesis as a path with a rugged terrain but under his guidance it was, to say it crudely, a cake walk and that’s the testament to his mentorship skills. I couldn’t have asked for a more perfect mentor and for that
you have my everlasting gratitude.
Lastly, I thoroughly enjoyed working on the research topic and I hope you enjoy reading it as
well😊
Table of Contents
ABSTRACT ... 5
INTRODUCTION ... 5
LITERATURE REVIEW ... 7
METHODOLOGY ... 16
STUDY 1 - UNEXPLAINED LOYALTY AND ITS IMPACT ON BEHAVIORAL LOYALTY ... 19
METHODOLOGY ... 20
MEASURE ... 21
DISCUSSION ... 25
STUDY 2a - LONGITUDINAL ANALYSIS OF SWITCHING COST ON CONSUMER BEHAVIOR ... 26
METHODOLOGY ... 26
RESULTS... 28
STUDY 2b - PREDICTING THE BEHAVIORS OF NON-SURVEY CONSUMERS ... 31
METHODOLOGY ... 31
RESULTS... 37
DISCUSSION ... 38
GENERAL DISCUSSION ... 39
LIMITATIONS OF THE PRESENT STUDY ... 40
APPENDIX - A ... 41
TO SWITCH OR NOT TO SWITCH?
A NEW APPROACH TO SWITCHING COSTS
ABSTRACT
Extant researchers collectively agree that consumer switching costs form an important defensive marketing tool for customer retention. However, there is surprisingly little agreement concerning the conceptualization and measurement of switching costs. This study illustrates the importance of measuring switching costs that is easy to estimate and can be readily used to infer an individual’s switching costs. The study uses the customer satisfaction survey from a financial service firm to create the switching cost measure and subsequently study the effect of the proposed switching cost measure on consumer behavior. Switching costs can be viewed as an unexplained customer loyalty that is the part of the loyalty that is not explained by customer satisfaction. Identifying consumers who are more vulnerable to switch has been an age-old battle in the CRM world. The proposed switching cost measure helps in identifying consumers who are vulnerable to switch before they gradually start depreciating their investments with an incumbent firm over time. Managers can then allocate resources to counteract the depreciating investments with the firm, thus evading customer attrition.
Keywords – Customer satisfaction, switching costs, attitudinal loyalty, behavioral loyalty,
consumer behavior
INTRODUCTION
The core ingredient of a firm’s marketing activity is focused on developing and enhancing
customer loyalty towards its products or services (Dick and Basu, 1994). The direct consequence of enhanced customer loyalty is increased revenue, lower acquisition cost, greater profitability, and reduced costs of serving repeat purchasers (Lam, Shankar and Murthy, 2004). Unsurprisingly, the classic customer loyalty-customer satisfaction framework has received an undivided attention in the marketing literature for close to 60 years (Watson, Beck, Henderson and Palmatier, 2015).
increase, the strength of relationship between repurchase intention and satisfaction diminish. Burnham, Frels and Mahajan (2003) posit switching costs as a multi-dimensional construct and demonstrate how greater switching costs are associated with higher customer intention to stay with an incumbent provider. Bansal and Taylor (1999) develop an empirical model on how switching costs predict switching intentions and switching behavior. The study highlights the fact that switching costs play a more important role in predicting switching intentions than customer satisfaction. Mittal (2016) acknowledges the possibility of customers switching service providers despite being satisfied and alternatively customers staying loyal to a service provider despite being unsatisfied.
Although switching costs function as a powerful defensive marketing tool by binding consumers to a firm and thus deterring consumers to switch from one supplier to another (Chebat, Davidow and Borges, 2010), switching cost studies form only an iota in the plethora of marketing literatures on customer loyalty/customer retention (Burnham, Frels, Mahajan,2003). Burnham, Frels and Mahajan (2003) notes that switching costs have largely been supported by anecdotes, logic and simplistic measures, therefore there is a need for creating a switching cost measure that is readily available and that can predict customer behaviors more accurately keeping in mind the relationship between customer satisfaction and customer loyalty.
Hardie, 2018) that predict the customer churn based on consumer behavior attributes. However, churn models predict customer churn based on “relationship diminishment” i.e. reduced frequency
of purchase/visits or non-purchase of products/services. The reversal of “relationship
diminishment” can come at a considerable cost to the firm. The proposed switching cost measure
allows managers to identify “bad apples in a basket” before the said “relationship diminishment”
occurs and therefore can adopt early counter-measures to avoid customer churn. The subsequent sections provide a brief overview on switching costs, existing switching cost measures and the formulation of the proposed switching cost measure based on the customer satisfaction-customer loyalty relationship and its practical implications.
LITERATURE REVIEW
SWITCHING COST – THE DEFINITION
Early definition of switching costs was coined by Porter (1980) who defines switching cost as a “one-time” cost as opposed to costs associated with using a product or a service. Burnham, Frels
and Mahajan (2003) add to this definition by defining “switching costs as “one time” costs that customers associate with the process of switching from one provider to another”.
Although, there are several definitions of switching costs conceptualized in several literatures, switching costs can be broadly classified into two categories – Objective monetary switching costs and non-monetary switching costs (Pick and Eisend, 1994).
Objective monetary switching cost
intends to change the operating system. Some of the most commonly recognized types of monetary switching costs are contractual costs, economic risk costs and monetary loss costs.
Contractual costs are the “costs that are present because of specific firm efforts or contracts
requiring the customer to pay a penalty to switch providers” (Guiltinan, 1989). Economic risk costs
are related to the risks involved in adopting a new provider that could potentially result in a negative outcome for a consumer (Burnham et.al, 2003). Finally, Monetary loss costs are one-time investments that are incurred by a consumer while switching to a new provider apart from the cost of buying a new product (Heide and Weiss, 1995).
Non-monetary switching cost
In this research stream, switching costs are individual specific perceptions of time, effort and emotions associated with changing the service providers (Jones et.al, 2000). Non-monetary switching costs arise when consumers make specific investments on a relationship with a supplier, and over time, the customer may have developed routines and procedures for dealing with the supplier (Lam et.al, 2004). Some of the common non-monetary switching costs are – search costs, learning costs and relational costs.
Search costs (Jones et.al, 2000) results from geographic dispersion of product/service alternatives. Learning costs (Alba and Hutchinson, 1987) are the time and effort that a consumer must spend to
familiarize with the new product or a service. Relational costs (Burnham et.al, 2003) are any psychological and emotional discomfort that arises from switching from one supplier to another.
EXISTING SWITCHING COST MEASURES AND DRAWBACKS
A wide range of switching cost measures have been used in prior research. Switching costs have largely been supported by anecdotes, logic, simplistic measures (Burnham et.al, 2003) and analytical models. The subsequent section highlights some of the common switching cost measures and their most evident drawbacks (refer Excerpt 1 for the summary).
Analytical Models
There are several studies in the marketing/economics literature that analyze the impact of switching costs in the market. The conceptualization of switching costs adopt various game-theoretic models and is analyzed either across two time periods or in an infinite time horizon. Beggs and Klemperer (1992) analyze the impact of switching costs on competition in a market across an infinite time horizon. They present the model at infinitely discrete time periods by analyzing two firms and how consumers switch across these two firms at various market situations i.e. a situation where one firm is a new entrant, the evolution of market share with either of the two firms being a market leader, change in market share in the absence of switching costs etc. However, the assumption of an infinite time-period is considered too unrealistic and therefore a model with a shorter buyer time horizon is necessary (To, 1995).
Direct Observation
Studies under this umbrella formulate switching costs based on the consumer switching behavior. For example, Dube, Hitsch and Rossi (2009) formulate switching costs based on brand loyalty and price to compute probabilities of switching to rival firms. The study captures the consumer heterogeneity and the effect of switching costs on competition at a firm level well. However, the study has an important assumption i.e. it assumes switching cost as a psychological construct rather than a complex amalgamation of monetary and non-monetary parameters as posited by Burnham et.al (2003). Cabral (2008); Shaffer and Zhang (2000) also point to the fact that switching behavior does not fully explain switching costs and there could be other alternative explanations such as aggressive pricing by firms both to existing and rival customers.
Indirect Estimation
using complex set of parameters and assumptions that has strong theoretical relevance but little impact on day to day managerial decision making.
Direct survey measurement
The most common method of carving out switching cost perceptions among individual consumers by market researchers. Bansal and Taylor (1999) create a switching cost measure that reflect an individual’s perception about access to resources and opportunities as well as his or her
self-assurance for engaging in a behavior of switching. They construct the switching cost measure using a cross-sectional consumer survey.
Ping (1993) constructs switching cost as the perceived magnitude of the cost and effort that would be required to change the supplier. The survey includes items that captures dimensions such as spending time and money, the benefits, and losses in switching suppliers. Finally, Burnham, Frels and Mahajan (2003) encapsulate perceptions of the cost involved in switching using a 5-point Likert scale. The survey is designed in such a way that an individual’s perceptions about
procedural, financial and relational switching costs are documented well. Cross-sectional direct survey measurements are excellent resources with high academic relevance although with limited managerial relevance. First, conducting cross-sectional surveys are expensive and can therefore not be conducted on a frequent basis to see the variations in consumer switching perceptions that can eventually be used in target marketing. Second, understanding the way switching costs affect subsequent consumer behaviors in a particular firm is also important and can only be done with a
To summarize, switching costs measures are quite diverse with different theoretical constructs, however there are procedural deficiencies in terms of generalizability and practical application. (Farrell and Klemperer 2007; Jones et al. 2002; Shy 2002).
Excerpt 1 – Summary of switching cost measures PREVIOUS
STUDY
CONCEPTUALIZATION OF SWITCHING COSTS
MEASURE USED
(TYPES OF COSTS CAPTURED)
ADVANTAGES/DISADVANTAGES
Beggs and Klemperer (1992)
Transport cost Analytical model that assumes switching cost as a transport cost to switch from firm A to firm B and
viewed at infinite time periods
Measures the effect of switching cost at different market scenarios
Assumptions are quite unrealistic
Villas-Boas (2011)
Termination, search and evaluation,
adoption costs
Two-period and Two firm model extended to an infinite time horizon
Captures effect of switching costs on price levels in such a single-price regime
Firms are assumed not to be able to price differentiate between new and existing customers
No consumer heterogeneity Dube, Hitsch and Rossi (2009) Psychological switching cost
Brand loyalty and price are varied to study the effect of switching costs on switching behavior
Switching costs, price, and brand loyalty are used to estimate consumer utility
Switching costs are not inherently psychological
Shy (2002) Function of brand price and market
share
Switching costs are estimated using Nash-Bertrand equilibrium concept
Quick and easy calculation of switching cost
Does not incorporate customer heterogeneity
Honka (2014)
Combination of demographic and psychographic factors
Combines firm-level and consumer level data to estimate customer inertia i.e. a combination of
satisfaction and switching costs
Formulates the impact of switching cost at firm-level and at consumer level
Complex assumptions made that are quite unrealistic
Burnham, Frels and Mahajan (2003) Multidimensional construct of switching cost
Cross sectional survey capturing consumer’s perception of switching costs based on procedural,
financial and relational switching costs
Elegant conceptualization of switching costs with high academic relevance
Has limited scope in terms of managerial decision making
Ping (1993) Perceived sacrifices consumers feel they will incur in switching
providers
Cross-sectional survey includes items that captures dimensions such as spending time and money, the
benefits, and losses in switching suppliers
Direct measurement of consumer perceptions
A new switching cost measure - the Loyalty, Satisfaction and Switching costs Triad
The concept of loyalty in marketing research has been formulated using diverse theoretical and operational approaches, with researchers selectively examining loyalty in terms of attitude but ignoring loyalty in terms of behavior or repeat patronage (Dick and Basu,1994, Reinartz and Kumar, 2002, Watson et.al, 2015). However, extant researchers agree that loyalty is a mixture of attitudes and behavior that benefit one firm relative to its competitors (Watson et.al, 2015).
Attitudinal loyalty is posited as a psychological construct that measures a consumer’s attitude
towards a brand or a firm (Fournier, 1998).
Alternatively, Behavioral Loyalty is defined as the measurement of a consumer’s past purchases of a particular brand or a brand set and predicting the probability of future purchases based on past purchase behaviors (Evanschitzky et. al, 2006). Consequently, loyalty can be viewed as a multi-dimensional construct with satisfaction playing an important role as a predictor of loyalty (Garbarino and Johnson, 1999). It is well known that satisfaction leads to favorable attitude towards a firm (attitudinal loyalty) and subsequently to repeat patronage (behavioral loyalty) and willingness to pay price premium for the shopping experience (Chaudhuri and Holbrook, 2001). Furthermore, switching costs are observed when consumers exhibit loyalty that cannot be explained by satisfaction delivered to them by the firm’s product or service offerings. (e.g. Fornell,
1992). Therefore, we construct the switching cost measure as “unexplained loyalty” while regressing the explanatory variable – Customer Satisfaction with the attitudinal measure of loyalty. The degree of “unexplained loyalty” is inversely proportional to the magnitude of attitudinal
Figure 1 – Loyalty -Satisfaction regression with error term
Therefore, the switching measure – unexplained loyalty forms the stochastic error component in the regression between Loyalty and Customer Satisfaction. Unexplained loyalty is a consumer specific measure that is computed using customer satisfaction survey data. The error term can be broadly classified into three terms i.e. positive residual, zero residual and negative residual respectively. Dick and Basu (1994) posit that consumers exhibit repeat patronage behavior despite having negative attitudes, which is usually temporary (e.g. lack of funds to afford other alternatives when consumption in a product class is considered necessary).
Figure 2 – Residuals classification for Loyalty-Satisfaction regression.
This behavior can be attributed to the class of cohorts who exhibit negative residuals i.e. individuals with negative attitudes in the satisfaction- loyalty relationship (refer figure 2). The group of consumers are classified as “prisoners” that is consumers who exhibit negative attitude
towards a firm but are bound to the firm due to situational constraints (e.g. lack of funds, lack of relevant alternatives) (Curasi & Kennedy, 2002). In the long run, prisoners would gradually scale down their investments towards the firm thus exhibiting “relationship diminishment” and overcoming the situational constraint that binds them to the firm.
Therefore –
H1. Prisoners would gradually depreciate their investments over time as compared to other
customer groups.
Beggs and Klemperer (1992) posit that in the presence of switching costs, certain rational customers engage with a firm while considering the costs and benefits of making the purchase decision before exhibiting repeat purchase behavior from the same firm in the future. This can be attributed to the “zero residual” customers who make an informed purchase decision with
switching cost playing a subdued role in their purchase decision.
The class of the customer cohort whose satisfaction perfectly explains their attitudinal loyalty can be classified as rational stayers. Over a finite time-period, rational stayers would increase their investments towards a firm albeit with a strong monetary and non-monetary motive. Therefore –
H2 Rational stayers would increase their investments but at a slower pace than positive residual
consumers
Finally, the group of consumers who have positive residuals exhibit higher levels of attitudinal
loyalty for a given satisfaction value. The group of consumers can, therefore, be classified as positive stayers (consumers who are happy and exhibit positive switching costs). Aaker (1996);
loyalty. Therefore, positive stayers should have highest growth in investments towards a firm over a given time-period as compared to the rest of the customer cohorts.
H3 Positive stayers should have the highest growth of investments towards a firm as compared to
the other customer cohorts
METHODOLOGY
DATA DESCRIPTION
Our data consists of a consumer survey data and a behavioral transaction data provided by a Fortune 100 financial services company in the United States. The behavioral transaction data include consumer specific variables such as revenue, number of products bought/owned and relationship tenure which were collected during a 13-month period in 2007-2008.
Table I. Sample profile (n = 10,000)
Demographic Per cent
Age (years) 18-24 4.1 25-34 12.5 35-44 22.9 45-54 32.0 55+ 28.4 Total 100 Gender Female 43.2 Male 56.8 Total 100 Household Income < US$15,000 7.6 US$15,000 - US$30,000 14.3 US$30,000 - US$50,000 23.5 US$50,000 - US$75,000 22.5 US$75,000 – US$100,000 16.2 > US$100,000 15.9 Total 100
UNEXPLAINED ATTITUDINAL LOYALTY MEASURE
We estimated the “unexplained attitudinal loyalty” using customer survey data of a financial service company. The overall satisfaction of the customer is “an overall evaluation on the total
Both the scales have been tested for construct validity and nomological validity by the respective authors. The test for the nomological validity of the proposed switching cost measure is presented in Appendix A. A First-order Auto-correlation, generalized least squares regression model is estimated with Loyalty (M = 6.17, SD = 3.16) as the dependent variable and Overall Satisfaction (M = 6.93, SD = 2.52) as the explanatory variable. The regression model parameters of a consumer
i is given by Equation 1.
Loyalty(i) = 0 + 1Satisfaction(i) + (i) (Eq. 1)
Table II – Results of regression analysis based on Equation 1.
Variable Estimate Standard Error t-statistic p-value
Intercept .79 .08 9.50 .00
Overall Satisfactiona .78 .01 70.98 .00
a - significant (p <.05)
A significant regression model is estimated where overall satisfaction (β1 - .78, p < .05) has a
positive impact on loyalty. The residuals estimated from the regression model are conceptualized as “unexplained attitudinal loyalty”. Based on the estimated “unexplained attitudinal loyalty”, we
Table III – Profile of consumers based on exhibited switching costs
Profile Positive Stayers Rational Stayers Prisoners Age (total – 100) 18-24 10% 6% 5% 25-34 18% 18% 18% 35-44 22% 24% 23% 45-54 23% 26% 24% 55+ 27% 26% 30% Gender (total - 100) Female 40% 46% 44% Male 60% 54% 56% Income (total – 100) <15,000 10% 6% 6% 15,000-30,000 18% 13% 13% 30,000 – 50,000 25% 23% 22% 50,000 – 75,000 21% 22% 26% 75,000 – 100,000 14% 18% 16% 100,000 > 12% 18% 17%
STUDY 1 - UNEXPLAINED LOYALTY AND ITS IMPACT ON BEHAVIORAL LOYALTY A CROSS SECTIONAL STUDY
switching costs and can therefore be targeted differently as compared to positive stayers who portray positive switching costs.
Ideally, it is plausible to test the influence of switching cost on relationship indicators by computing the switching cost measure at different time periods and measuring the influence on relationship indicators at each time-period. However, due to data limitations and for the sake of simplicity, we assume switching costs as a stationary ergodic process where the mean values, moments and variance are constant over time. Furthermore, Bernhardt, Donthu and Kennett (1999) posit that the effect of customer satisfaction does not significantly improve an incumbent firm’s profitability at shorter time periods. There is, however, a positive significant effect of customer satisfaction in the long run. Anderson et. al, (1994) also propose that customer satisfaction has a positive effect on market share and profitability of an incumbent firm at future time periods. The study is conducted over a five-year time period as compared to the present study where consumer behavior is analyzed over a thirteen-month time-period. Since, the switching costs measure created in the present study is a function of customer satisfaction, we expect the switching cost measure to be stable over the thirteen-month time-period, particularly, for a financial services company.
METHODOLOGY
A sample of 10,000 customer from a Fortune 100 financial services company undertook the customer satisfaction survey. We measure the “unexplained attitudinal loyalty” for the given
(Lam et.al ,2004; Chebat, Davidow, Borges; 2010). Furthermore, the relationship between the number of products owned and monthly deposits depends on the type of switching costs exhibited by the consumer. For example, positive stayers would exhibit higher growth of investments as compared to prisoners who would eventually scale down their investments. Therefore, switching costs should play the role of a moderator. Consumer Trust is another important determinant of behavioral loyalty (Watson et.al, 2015) and is defined as a the “confidence in the reliability and integrity of a seller (Watson et.al, 2015). Trust is constructed on a 10-point Likert scale as posited by Watson et.al (2015). The depth of investments depends on the level of trust that each consumer has towards the firm and therefore Trust should moderate the influence between monthly deposits and number of products owned.
MEASURE
A total of 7,571 customers are included in the study after the treatment for outliers and missing
values without significantly changing the sample profile. We use the logarithmic-transformation
of monthly deposits and standardize the variable trust (M = 0, SD = 1) to accommodate for the right skewed nature of the variables without complicating the model structure (Mallapragada, Chandukala and Liu, 2016).Table IV shows the descriptive statistics of the selected variables.
Table IV – Descriptive statistics and Construct Intercorrelations (N = 7,571)
Variable Mean Standard Deviation
1 2 3 4
1. Products Added .60 .80 1.00
2. Monthly deposita 2.86 3.85 .43 1.00
3. Unexplained Attitudinal Loyalty 0.00 2.30 .07 .05 1.00
4. Consumer Trustb 0 1 .11 .05 .32 1.00
To analyze the moderating role of switching costs and trust, we ran a regression that tested the main effect of monthly deposits, switching costs and trust and the interaction of switching cost and monthly deposits and the interaction of trust and monthly deposits on number of products owned/purchased. We estimate the moderated regression model by using an aggregate data (mean values of each consumer variable). The main purpose of the study is to test the influence of switching costs on future consumer investments and therefore we conduct a cross-sectional study.
The moderated regression is performed using process macro to compute Johnson-Neyman points (Hayes, 2012). The computation of Johnson-Neyman points also referred to as Floodlight analysis highlights the entire range in a continuous variable X where the simple effect in moderated regression is significant and not significant (Spiller, Fitzsimons, Lynch, and McClelland, 2013). This is particularly relevant to our study since we are interested in how positive, zero and negative switching costs influence future investments at different levels of trust.
RESULTS
Table V presents the main effects and the interaction effects of the moderated regression model.
Table V – Results – Moderating effects of switching costs and Trust in Products added and Monthly deposits link
Model Estimate Standard Error Significance
A significant regression model was estimated with the predictors explaining 45 per cent of the variation in number of products added (R2 = .45, F (5, 7.565) = 385.13, p < .05). All the parameter
estimates are significant at 5 per cent significance level except for switching costs (β - .003, p > .05). However, the interaction effect between monthly deposits and switching costs (β - .002, p < .05) have a positive significant effect on number of products added. Figure 3 illustrates the moderating role of switching costs in the products added - monthly deposits link.
Figure 3 – Moderating role of switching costs in the Products added – Monthly deposits link
Similarly, the interaction effect between monthly deposits and consumer trust also has a positive significant (β - .007, p < .05) on number of products added. The effect of consumer trust on products added is positive and significant (β - .044, p < .05). This result is analogous to the effect of consumer trust on behavioral loyalty described in Watson et.al (2015). Figure 4 illustrates the interaction between trust and the monthly deposit-products added relationship.
Pr o d u cts ad d e d Monthly deposits
Figure 4 – Moderating role of consumer trust in the Products added – Monthly deposits link
We compute the Johnson-Neyman points of the interaction terms to analyze the effect on products purchased at levels of consumer trust and switching costs. Table VI describes the results of the floodlight analysis.
Table VI. Conditional effects of the focal predictor – products added at values of the moderator
Consumer Trust Switching costs Effect Standard Error Significance value
-1.013 -2.070 .0761 .003 .000 -1.013 .284 .0820 .003 .000 -1.013 1.929 .0861 .004 .000 .260 -2.070 .0862 .003 .000 .260 .284 .0921 .002 .000 .260 1.929 .0961 .002 .000 1.121 -2.070 .0929 .004 .000 1.121 .284 .0988 .003 .000 1.121 1.929 .1029 .003 .000
Note - Johnson-Neyman points are estimated at 16th,50th, and 84th percentile of the moderator variables
Pr o d u cts ad d e d Monthly deposit
The result of the floodlight analysis shows that the effect of consumer trust on number of products added is significantly higher (p < .05) at higher levels of trust. Similarly, positive switching cost (positive stayers) has the highest effect on the number of products added, followed by zero residual switching cost (rational stayers) and negative switching costs (prisoners) respectively. The results are analogous to Jones et.al (2007) who posit that positive and negative switching costs have a divergent effect on repeat purchase intentions.
DISCUSSION
product purchases. The study mimics the results posited by Watson et.al (2015) where higher levels of consumer trust have positive influence on future product purchases (behavior loyalty).
STUDY 2a - LONGITUDINAL ANALYSIS OF SWITCHING COST ON CONSUMER BEHAVIOR
The cross-sectional study of the effect of switching costs on future product purchases does not capture the “relationship diminishment” behavior of consumers. Furthermore, analogous to our study, Bernhadt et.al (2000) posit that cross-sectional studies with respect to the effect of customer satisfaction on profitability have found inconclusive evidence and highlight the importance of a longitudinal analysis of customer satisfaction. Lee et.al (2001) posit the importance of tracking consumption behavior to infer behavioral switching cost for firms providing continuous services (e.g. banks, mobile operators, insurance industry etc.). Therefore, in this study, we test for the formulated hypothesis by tracking consumer behavior across time using a Bayesian multilevel model.
METHODOLOGY
A total of 7,571 customers are included in the study after the treatment for outliers and missing
values without significantly changing the sample profile. We replicate the variables in study 1 for
Table VII – Hypothesis to be tested
Hypothesis framework
Hypothesis I Prisoners would gradually depreciate their investments over time as compared to other customer groups
Hypothesis II Rational stayers would increase their investments but at a slower pace than positive residual consumers
Hypothesis III Positive stayers should have the highest growth of investments towards a firm as compared to the other customer cohorts
Hierarchical Bayes procedure is quite prominent in marketing literatures due to the computational and modeling advancements made in recent times (Rossi and Allenby, 2014). Furthermore, given the longitudinal nature of marketing data, past literatures on customer satisfaction (Larivere, Keiningham, Aksoy, Yalcin, Morgeson, and Mithas, 2016) use Bayesian estimation methods to account for consumer heterogeneity, which forms the fulcrum of targeted marketing actions. Therefore, we adopt the hierarchical bayes procedure using Markov chain Monte Carlo simulation method (Rossi and Allenby, 2014) in the present study. We run four independent Markov chain Monte Carlo (MCMC) chains with an uninformed prior (default) and 4,000 iterations per chain. The first 1,000 iterations are considered the “burn-in” phase and the remaining 3,000 iterations are used to estimate the posterior distribution for the parameters, resulting in a distribution based on 12,000 points. To test for the convergence of the MCMC algorithm, we use the Gelman-Rubin convergence statistic (Rhat) as posited by Gelman and Rubin (1992).
The model framework for the Bayesian multi-level model for a consumer i and time-period t is specified in Equation 1. The parameters γ1, γ2 are the population-averaged (fixed effect) slope
coefficients with a population-averaged intercept γ0, whereas u1i and u2i are the group-level
(random effect) slope coefficients that vary among consumers with a consumer-specific intercept u0i.
Level 1 - Products added(it) = β0i + β1i time period(it) + β2i monthly deposit(it) + β3Trust(i) + (it) (Eq. 1)
Level 2 - β0i = γ0 + u0i (Eq. 2.1)
Level 2 - β1i = γ 1 + u1i (Eq. 2.2)
Level 2 - β2i = γ 2 + u2i (Eq. 2.3)
RESULTS
Table VIII shows the results of the population level estimate of the Bayesian multilevel model. All the independent variables have the Gelman-Rubin convergence statistic (Rhat) equal to 1.0 which suggests that the four independent chains have converged and that additional iterations and setting stronger priors is not necessary. Additionally, the mean of the posterior distribution for time suggests that Positive stayers have a slight growth in future product purchase (γ1 - .002). This
confirms Hypothesis III where consumers having strong attitude and exhibiting positive switching cost are likely to make greater investments towards the firm. Alternatively, Prisoners have a degrowth in terms of product purchase (γ1 - -.013) which in turn confirms Hypothesis II.
Theoretically, prisoners are consumers who have negative attitude towards the firm and are likely to scale down their investments over time. Furthermore, Rational stayers exhibit a slight degrowth in terms of product purchase (γ1 - -.008) and therefore we fail to confirm Hypothesis I. However,
are involved in rational decision making, therefore these consumers could show degrowth in terms of product purchase due to lack of attractive product offering. The 95% Bayesian confidence intervals give the 2.5 and 97.5 percentiles in the posterior distribution. The confidence interval for the time estimates does not include zero thus suggesting a significant effect of time on product purchase. Furthermore, higher levels of consumer trust have a positive impact on future product purchase which is a replication of Study 1.
Table VIII – Population-level (Fixed) effects of Bayesian Multi-level model
Estimate Posterior SD 95% Confidence
Interval R-hat Positive Stayers Intercept (γ0) .760 .040 (.68, .83) 1.00 Time (γ1) .002 .003 (.00, .01) 1.00 Monthly depositsa (γ 2) .052 .003 (.05, .06) 1.00 Trustb (β 3) .051 .030 (.00, .11) 1.00 Rational stayers Intercept (γ0) .820 .042 (.74, .90) 1.00 Time (γ1) - .008 .003 (-.02, .00) 1.00 Monthly depositsa (γ 2) .042 .003 (.03, .05) 1.00 Trustb (β 3) .076 .003 (.02, .14) 1.00 Prisoners Intercept (γ0) .750 .042 (.67, .83) 1.00 Time (γ1) -.013 .003 (-.02, -.01) 1.00 Monthly depositsa (γ 2) .042 .004 (.03, .05) 1.00 Trustb (β 3) .050 .030 (-.01, .11) 1.00
Notes: Posterior SD – Posterior Standard Deviation.
Table IX shows the random effects of the estimated Bayesian multi-level model. The estimation of random intercepts allows different consumers to have different intercepts. Furthermore, we estimated the random coefficients for time (u1i) and monthly deposit (u2i) to account for consumer
heterogeneity. Thus, each consumer can have his/her own coefficient to estimate product purchase, which is the sum of the population average and the deviation from it (population-level effect (γ) + deviation (ui)). From Table IX, it can be inferred that the growth of future product purchase is
higher for consumers with high monthly deposits for all customer cohorts i.e. Positive stayers, Rational stayers, and Prisoners.
Table IX – Group-level (Random) effects of Bayesian Multi-level model
Estimate Posterior SD 95% Confidence
Interval R-hat Positive Stayers Intercept (u0i) 1.090 .03 (1.03, 1.15) 1.00 Time (u1i) .110 .00 (.10, .12) 1.00 Monthly deposits (u2i) .080 .00 (.07, .08) 1.00 Rational stayers Intercept (u0i) 1.120 .04 (1.05, 1.19) 1.00 Time (u1i) .110 .00 (.10, .11) 1.00 Monthly deposits (u2i) .080 .00 (.07, .08) 1.00 Prisoners Intercept (u0i) 1.050 .03 (.98, 1.12) 1.00 Time (u1i) .080 .00 (.07, .09) 1.00 Monthly deposits (u2i) .100 .00 (.09, .11) 1.00
STUDY 2b - PREDICTING THE BEHAVIORS OF NON-SURVEY CONSUMERS USING PROPENSITY SCORE MATCHING
Study 2a highlights the importance of switching cost measure and its practical application of tracking “relationship diminishment” behavior of consumers in future time periods. However,
study 2a is analyzed for a sample of consumers who undertook the customer satisfaction survey for a financial services company. We extend the analysis in Study 2a to non-survey consumers by implementing a matching procedure using Propensity score matching. Although, Propensity score matching procedures are widely used in observational or quasi-experimental studies to reduce selection bias (Kim, Wang, and Malthouse, 2015; Garnefeld, Eggert, Helm and Tax, 2013), the main premise behind using the procedure is in its application. Propensity score matching procedure matches each consumer who participated in the customer satisfaction survey with a similar customer (“statistical twin”) who did not participate based on a defined set of covariates (Garnefeld
et.al, 2015; Rosenbaum and Rubin, 1983). In the subsequent sections, we first match the consumers who undertook the survey with the non-participant consumers. Second, we check for the quality of matching. Third, we infer the switching costs incurred by non-participants based on switching costs exhibited by their respective “statistical twin” as described earlier. Finally, we estimate the growth model of non-participants based on the exhibited switching costs, which is similar to study 2a (conducted for survey participants).
METHODOLOGY
A total of 2,799 consumers who participated in the customer satisfaction survey (treatment group) are included in the study after the treatment for outliers and missing values. Correspondingly, we include 44,059 non-participants (control group) in the study after fixing for outliers and missing
Table X – Descriptive statistics and construct intercorrelations between treatment and control group
Variable Mean Standard
Deviation
1 2 3 4 5
Participants
(Treatment group)
1. Net interest income 75.18 110.37 1.00
2. Tenure (in months) 125.30 104.95 .178 1.00
3. Age 47.13 13.51 .040 .349 1.00
4. Sales cost 6.31 9.46 .412 .035 .349 1.00
5. Net interest expense 41.93 34.59 .462 .241 .951 -.006 .730
Non-participants
(Control group)
1. Net interest income 56.45 103.45 1.00
2. Tenure (in months) 135.13 100.69 .154 1.00
3. Age 53.40 14.53 -.020 .286 1.00
4. Sales cost 4.16 7.84 .340 .052 -.006 1.00
5. Net interest expense 21.00 29.47 .463 .240 -.003 .735 1.00
Table XI – Testing for imbalance before matching
Estimate Chi-square statistic Degrees of freedom p-value
Unstrat 462 2 .00**
** - significant (p <.01)
The chi-square test suggests that the difference between the two groups are indeed significant (p
< .01). However, we still use Age and Tenure for our matching procedure and test for the balance
between samples after the matching procedure is conducted to check their relevance.
Propensity score matching
RESULTS
Table XII shows the group means before and after matching and the percentage reduction in bias. Percentage reduction in bias is calculated for all variables as posited by Rosenbaum and Rubin (1983).
Table XII – Group Means Before and After Matching and Percentage Reduction in Bias
Before Matching After Matching
Control group (N = 44,059)
Treatment group (N = 2,799)
Covariates Control Group (N = 2,799) Treatment Group (N = 2,799) Reduction in Percentage Biasa 53.40 47.13 Age 47.17 47.13 .99
135.13 125.30 Tenure (in months) 129.79 125.30 .54
a – Percentage reduction in bias is calculated using the formula posited by Rosenbaum and Rubin (1983)
Formula for percentage reduction in bias (Rosenbaum and Rubin, 1983) -
PRB = 1- | 𝑋𝑖
𝐴− 𝑋 𝑗𝐴 |
| 𝑋𝑖𝐵− 𝑋𝑗𝐵 |
PRB – percentage reduction in bias
𝑋
𝑖𝐴–
the mean for the treatment group after matching𝑋
𝑗𝐴–
the mean for the control group after matching𝑋
𝑖𝐵–
the mean for the treatment group before matching𝑋
𝑗𝐵–
the mean for the control group before matchingtest for the quality of matching procedure based on the omnibus test as posited by Hansen and Bowers (2008). Table XIII results suggest that the difference between the participant and the non-participant groups are not significant at the 5 percent significance level. Therefore, the covariates used in the study are relevant for the matching procedure.
Table XIII – Testing for imbalance after matching
Estimate Chi-square statistic Degrees of freedom p-value
Unstrata 1.51 2 .471
a - significant if (p <.05)
Figure 5 illustrates the propensity score distribution between the matched treatment units and the control units. As can be seen in the figure, the section labeled ‘Unmatched control units’ shows that the participants are evenly distributed across the range of the propensity scores. After having established the quality of the matching procedure, we identify the individual “statistical twin” for the matched non-participant and infer the switching costs exhibited by the non-participant based on the switching costs exhibited by his/her matched counterpart (survey participant).
Figure 6 shows the distribution of switching costs exhibited by the matched non-participants. A total of 966 consumers are Rational stayers thereby exhibiting zero switching costs, followed by 829 prisoners who exhibit negative switching cost and 1,004 positive stayers exhibit positive switching costs respectively.
Figure 6 –Distribution of the switching costs exhibited by the non-participants
Growth model
We finally estimate a linear growth model for the non-matched participants based on the inferred switching costs. The dependent variable chosen for the growth model estimation is the same as Study 2a i.e. number of products purchased. Furthermore, the explanatory variable - time-period is incorporated as a fixed effect to measure the population-level trajectory of future product purchase growth. Alternatively, we estimate consumer specific intercepts (random effect) to account for consumer heterogeneity. The linear growth model is estimated using restricted maximum likelihood estimation (REML) as posited by Bates, Machler, Bolker, and Walker (2015). Ideally, we expect the switching cost behavior for non-participants to mimic the results presented in study 2a for the survey participants.
1,004 966 829 0 200 400 600 800 1000 1200
Positive stayers Rational stayer Prisoner
RESULTS
Table XIV illustrates the fixed effects for the linear growth model. The results from the growth model are starkly similar to the results estimated for the survey participants in Study 2a. Positive stayers have a slight growth in future product purchase (β – 7.23 X 10-6) as compared to Rational stayers who have a slight degrowth (β – -3.69 X 10-5) but lesser in magnitude as compared to
Prisoners (β – -3.77 X 10-5). We refrain from computing the p-values for the model primarily
because the assumption of a t-distribution for mixed effects model based on restricted maximum likelihood criterion is disputed (Bates et.al, 2015). More importantly, the theme of Study 2b is to illustrate the practical application of the switching cost measure and therefore, a linear growth model with restricted maximum likelihood criterion is easy to estimate with limited computational requirements.
Table XIV – Fixed effects for the Linear Growth model
Estimate Standard Error t-statistic
Table XV illustrates the random effects for the growth model. To account for consumer heterogeneity, each participant has his/her own intercept, which is the sum of the population-level intercept and its deviation from it.
Table XV – Random effects for the Linear Growth model
Variance Standard Deviation Positive Stayers Intercept .577 .760 Rational stayers Intercept .468 .684 Prisoners Intercept .484 .695 DISCUSSION
Tracking consumer behavior in terms of switching costs has wide theoretical implications. The
positive attitude towards the firm, show strong growth in product purchase. This is in line with previous study where Aaker (1996); (Evanschitzky, et.al, 2006) propose that the strength of consumer attitudes is a strong indicator of the loyalty behavior, with higher levels of attitudinal loyalty having positive impact on behavioral loyalty. Alternatively, Jones et.al (2007) propose that consumers exhibiting positive switching cost have a greater influence on repeat purchase intention. Study 2b illustrates a simple example where the switching cost measure can be extended to the non-survey participants. The results are quite similar to the results discussed above. We use unrelated variables in Study 2b (Tenure and Age) to infer switching costs for non-survey participants, thus overcoming any endogeneity problems that may arise in Study 2a with respect to the switching costs measure.
GENERAL DISCUSSION
consumers experiencing positive switching costs (we call them – positive stayers) make greater investments in future time periods in contrast with consumers experiencing negative switching costs (we call them - prisoners) who gradually depreciate investments over time. Furthermore, Hauser et.al (1994) highlight the need to track consumer behavior with the help of behavioral switching costs in continuing service firms. Study 2b provides a simple illustration of this
requirement. The results in Study 2b are estimated for non-survey participants with two variables that are unrelated to the switching cost measure, the results, however, are similar to the survey participants. Thus, Managers can use the switching cost measure created in this study as an
“early behavioral indicator” tool before the said “relationship diminishment” occurs and take
relevant counter measures to change the consumer attitude.
LIMITATIONS OF THE PRESENT STUDY
APPENDIX - A
INITIAL NOMOLOGICAL VALIDITY ASSESSMENT OF PROPOSED NEW MEASURE
VALIDITY ASSESSMENT RATIONALE VALIDITY ASSESSMENT RESULTS FOR NEW SWITCHING COSTS
MEASURE USING ACSIDATA OPERATIONALIZATION Due to the higher downside risk associated with purchasing
products of unknown quality, switching costs should be higher for durable than non-durable products.
The mean switching costs for durable products firms (0.09) is significantly (p < 0.05) higher than for those producing non-durable
(-0.11) products.
Since it is associated with brand choice, switching costs should be positively correlated with brand salience.
The correlation between switching costs and brand salience (EquiTrend©) for the firms in our data set is 0.16 (p < 0.05).
Switching costs should be higher for categories in which location can create quasi "local monopolies" (e.g., retail) than those where products/services may be supplied through multiple channels (e.g., food and beverage).
The mean switching costs in supermarket retailing (1.20) and discount retailing (0.95) are significantly higher (t-tests significant p < 0.05) than those for processed foods (0.36) and soft drinks (0.03).
Switching costs should be higher in categories well-known for high behavioral loyalty (e.g., cigarettes) vs. relatively lower behavioral loyalty (e.g., automobiles, apparel).
The mean switching costs in the cigarette category is 1.39 vs. a mean level of -1.74 for automobiles and -0.03 for apparel (t-tests significant at p < 0.05).
REFERENCE
Aaker D. (1996), “Building Strong Brands.” Boston, MA – The Free Press.
Alba, J. W. and Hutchinson, J.W (1987), “Dimensions of consumer expertise,” Journal of
Consumer Research, 13, 411-454
Anderson, E.W., Fornell, C. and Lehmann, D.R (1994), “Customer satisfaction, Market share and profitability – Findings from Sweden,” Journal of Market, 58, 33-66
Ascarza, E., Netzer, O., and Hardie, B.G.S (2018), “Some Consumers Would Rather Leave Without Saying GoodBye,” Marketing Science, 37(1), 54-77
Bansal, H.S. and Taylor, S.F. (1999), "The Service Provider Switching Model (SPSM): A Model of Consumer Switching Behavior in the Services Industry," Journal of Service Research, 21, 200-218.
Bates, D., Machler, M., Bolker, B.M., and Walker, S.C. (2015), “Fitting Linear Mixed-Effects Models using lme4”, Journal of Statistical Software, 67(1), 1-51.
Bernhardt, K.L., Donthu, N. and Kennett, P.A. (2000), “A Longitudinal Analysis of Satisfaction and Profitability,” Journal of Business Research, 47, 161-171
Beggs, A., and Klemperer, P.D. (1992), "Multiperiod Competition with Switching Costs,"
Econometrica, 60, 651- 666.
Burnham, T. A., Frels, J.K. and Mahajan, V., (2003), "Consumer Switching Costs: A Typology, Antecedents, and Consequences," Journal of the Academy of Marketing Science, 31(2) 109-126.
Cabral, L. (2008), "Switching Costs and Price Competition," working paper, Stern School of Business, New York University.
Chaudhuri, A. and Holbrook, M.B (2001), “The chain of effects from brand trust and brand effect to brand performance – the role of brand loyalty,” Journal of Marketing, 65, 81-93. Chebat, J.C., Davidow, M., Borges, A. (2011), “More on the role of switching costs in service
markets: A research note,” Journal of Business Research, 64, 823-829
Curasi, C.F., and Kennedy, K.N., (2002),“From prisoners to apostle: a typology of repeat buyers and loyal customers in service businesses,” Journal of Services Marketing, 16 (4), 322-341 Dick, A.S. and Basu, K., (1994), "Customer Loyalty: Toward an Integrated Conceptual
Framework," Journal of Academy of Marketing Science, 22, 99-113.
Dube, J.P., Hitsch, G.J., Rossi, P.E. (2009), “Do switching costs make markets less competitive?, ” Journal of Marketing Research, 46, 435-445
Evanschitzky, H., Iyer, G.R., Plassmann, H., Niessing, J. and Meffert, H., (2006), “The relative strength of affective commitment in securing loyalty in service relationships,” Journal of
Business Research, 59, 1207-1213
Farrell, J., and Klemperer,P., (2007), "Coordination and Lock- In: Competition with Switching costs and Network Effects," in M. Armstrong and R. Porter (Eds.), Handbook of Industrial
Fornell, C, (1992), "A National Customer Satisfaction Barometer: The Swedish Experience,"
Journal of Marketing, 56, 6-21.
Fournier S. (1998), “Consumers and their brands developing relationship theory in consumer research,” Journal of Consumer Research, 23, 343-73
Garbarino, E and Johnson, M.S (1999),“The different roles of satisfaction, trust and commitment in customer relationships” Journal of Marketing, 63, 70-87
Garnefeld, I., Eggert, A., Helm, S.V., Tax, S.S (2013), “Growing Existing Customers Revenue Streams Through Customer Referral Programs”, Journal of Marketing, 77, 17-32.
Gelman, A., and Rubin, D.B. (1992), “Inference from Iterative Simulation Using Multiple Sequences,” Statistical Science, 7(4), 457-72.
Guiltinan, J.P (1989), “A Classification of switching costs with implication for relationship market” AMA Winter Educators Conference: Marketing Theory and Practice. 216-20. Hansen, B.B., and Bowers, J. (2008), “Covariate balance in simple, stratified and clustered
comparative studies”, Statistical Science, 23(2), 219-236.
Hauser, J.R., Simester, D.I. and Wernerfelt, B. (1994), “Customer satisfaction incentives”,
Marketing Science, 13 (4), 327-50
Hayes, A. F. (2012). PROCESS: A versatile computational tool for observed variable mediation, moderation, and conditional process modeling [White paper]. Retrieved from
http://www.afhayes.com/ public/process2012.pdf
Heide, Jan B., and Allen M. Weiss (1995), “Vendor consideration and switching behavior for buyers in high technology markets,” Journal of Marketing, 59, 30-43
Honka, Elizabeth (2010), "Quantifying Search and Switching Costs in the U.S. Auto Insurance Industry," working paper, Jindal School of Management, University of Texas at Dallas. Ho, D.E., Imai, K., King, G., and Stuart, E.A. (2011), “ Matchit: Nonparametric Preprocessing
For Parametric Causal inference,” Journal of Statistical Software, 42(8), 1-28.
Jones, Michael A., Mothersbaugh, David L., and Beatty, Sharon E. (2000), "Switching Barriers and Repurchase Intentions in Services," Journal of Retailing, 76(2), 259-74.
Jones, Michael A., David L. Mothersbaugh, and Sharon E. Beatty (2002), "Why Customers Stay: Measuring the Underlying Dimensions of Service Switching Costs and Managing their
Differential Strategic Outcomes," Journal of Business Research, 55, 441-450.
Jones, Michael A., Kristy E. Reynolds, David L. Mothersbaugh, and Sharon E. Beatty (2007), "The Positive and Negative Effects of Switching Costs on Relational Outcomes," Journal of
Service Research, 9(4) 335-355.
Kim, S.J., Wang, R.H.J, and Malthouse, E.C., (2015), “The Effects of Adopting and Using a Brand’s Mobile Application on Consumers subsequent Purchase Behavior”, Journal of
Lam, Shun Yin, Venkatesh Shankar, M. Krishna Erramilli, and Bvsan Murthy (2004), "Customer Value, Satisfaction, Loyalty, and Switching Costs: An Illustration from a Business-to-Business Service Context," Journal of the Academy of Marketing Science, 32(3), 293-311.
Lariviere, B., Keiningham, T.L., Aksoy, L., Yalcin, A., Morgeson, F.V., and Mithas, S., “Modeling Heterogeneity in the Satisfaction, Loyalty intention, and Shareholder Value Linkage: A cross industry analysis at Customer and Firm Levels”, Journal of Marketing
Research, 53, 91-109.
Lee, J., Lee, J., Feick, L. (2001), “The impact of switching costs on the customer satisfaction-loyalty link: mobile phone service in France,” Journal of Services Marketing,15(1), 35-48 Mallapragada, G., Chandakula, S.R., and Liu, Q., (2016) “Exploring the Effects of What
(Product) and Where (Website) Characteristics on Online Shopping Behavior,” Journal of
Marketing, 80, 21-38
Maute, M.F., and William, R.F., Jr. (1993). “The Structure and Determinants of Consumer Complaint Intentions and Behavior,” Journal of Economic Psychology,14, 219-247 Mittal, B. (2016), “Retrospective: why do consumers switch? The dynamics of satisfaction
versus loyalty,” Journal of Business Research, 30(6), 569-575
Neslin, S.A., Gupta, S., Kamakura, W., Lu, J.and Mason, C.H. (2006), “Defection Detection: Measuring and Understanding the Predictive Accuracy of Customer Churn Models,” Journal
of Marketing Research,43, 204-211
Olmos, A., and Govindasamy, P. (2015), “Propensity scores – A practical introduction using R”,
Journal of Multidisciplinary Education, 11(25), 68-88
Porter, Michael E. (1980), Competitive Strategy, Free Press: New York.
Pick, D. and Eisend, M. (2014), “Buyers perceived switching costs and switching: a meta analytic assessment of their antecendents,” Journal of the Academy of Marketing Science, 42, 186-204
Ping, Robert (1993), "The Effects of Satisfaction and Structural Constraints on Retailer Exiting, Voice, Loyalty, Opportunism, and Neglect," Journal of Retailing, 69(3), 320-352.
Reinartz, W. and Kumar, V. (2002), “The mismanagement of customer loyalty”, Harvard
Business Review, 80, 86-94
Roos, Inger and Angers Gustafsson (2007), "Understanding Frequent Switching Patterns – A Crucial Element in Managing Customer Relationships," Journal of Service Research, 10(1), 93-108.
Roos, Inger and Angers Gustafsson (2011), "The Influence of Active and Passive Customer Behavior on Switching in Customer Relationships," Managing Service Quality, 21(5), 448-464.
Rosenbaum, P.R., and Rubin, D.B., (1983), “The Central Role of the Propensity Score in Observational Studies for Causal effects,” Biometrika, 70(1), 41-55
Rossi, P.E., and Allenby, G.M., (2003), “ Bayesian Statistics and Marketing”, Marketing
Shaffer, Greg and Z. John Zhang (2000), "Preference-Based Price Discrimination in Markets with Switching Costs," Journal of Economics & Management Strategy, 9(3), 397-424
Shy, Oz (2002), "A Quick-and Easy Method for Estimating Search Costs," International Journal
of Industrial Organization, 20, 71-87.
Spiller, A.S, Fitzsimons, G.S, Lynch, J.G and McClelland, G.H. (2013), “Spotlights, Floodlights, and the Magic Number Zero : Simple Effects Tests in Moderated Regression,” Journal of
Marketing Research, 12, 277-288
To, T (1996), “Multi-period competition with switching costs – an overlapping generations formulation,” The Journal of Industrial Economics, 44 (1), 81-87
Villas-Boas, J. Miguel (2011), "Notes on Switching Costs and Dynamic Competition," Working Paper, Haas School of Business, University of California at Berkeley.
