Stated Choice Probabilities and the Effect of Time
on Multi-Part Pricing in the Market of Mobile
Subscriptions
University of Groningen
Faculty of Economics and Business
MSc Marketing Thesis
June 20, 2016
1
stSupervisor: Dr. Keyvan Dehmamy
2
ndSupervisor: Dr. Felix Eggers
Hendrik Schulze Böing
S2001780
9711GE Groningen
0627097291
E-mail: k.h.schulze.boing@student.rug.nl
Stated Choice Probabilities and the Effect of Time on Multi-Part Pricing in the
Market of Mobile Subscriptions
Acknowledgment
I want to thank my supervisor Dr. Keyvan Dehmamy for the support and feedback throughout the whole process of writing this thesis.
Abstract
In this paper, elicited choice probabilities are used to investigate the effect of the time dimension on consumer preferences in the market of mobile subscriptions. A hierarchical Bayes model is used to estimate two sets of utilities, relating to two different moments in time. After showing that individual utilities change over time, two practical examples on the possible utilization of such data in the market of mobile subscriptions are presented: It can be shown that time-dynamic
preferences are well suited for the generation of optimal contract offers as well as for studying the evolution of between segment heterogeneity. Last but not least, the paper introduces time dynamics into the controversial field of multi part priced subscriptions. Herby, a trend towards flat- fee minutes and multi-part priced data volume is found.
Contents
Introduction ... 1 1 Research framework ... 3 Conceptual model ... 3 1.2 Dependent variable ... 5 1.3 Control variables ... 6 1.4 Attributes ... 6 1.5 Levels ... 7 1.6 Experimental Design ... 81.7 Choice set design ... 8
1.8 Factorial design ... 9
2. Methodology ... 9
3. Data ... 10
3.1 Estimation ... 12
3.1.1 Part worth specification ... 13
3.1.2 Linear specification ... 14
3.2 The effect of time on individual utilities ... 17
3.2.1 The effect of time important marketing variables at the individual level ... 18
3.2.2 The effect of time on uncertainty ... 18
5. Discussion ... 19
5.1 Multi part pricing ... 19
5.2 Optimal subscription offers at the individual level ... 21
5.3 Segmentation ... 23
5.4 Limitations ... 25
6. Conclusion ... 26
Reference ... 27
1
Introduction
Choice based conjoint has enabled marketers to better understand and measure customer preferences, ever since its introduction in the 1980s. Moreover, conjoint analysis seems to be constantly evolving and one might argue that “new developments in conjoint analysis are arriving so fast that even specialists find it difficult to keep up” (Carroll and Green 1995). The most important developments after the initial introduction by Louviere and Woodworth (1983) include the usage of Hierarchical Bayes as well as latent class modelling (Agarwal 2014).
However, the development of choice based conjoint (CBC) is far from over. According to the meta-analytic work by Agarwal (2014), time-discounted utility models and the inclusion of behavioral effects make up an important direction for future research. Respondent uncertainty plays a crucial role in these new approaches and efforts have been made to enable its measurement. The works of Manski (2004) and Asher et al. (2010) make a convincing case in favor of the elicitation of choice probabilities over the elicitation of discrete choices in order to account for respondent uncertainty.
Time discounted utility models or time-dynamic choice based conjoint are desirable because they represent a more “realistic” setting. One can imagine that the time of purchase has an impact on the utility of a good or service and that this in turn effects the optimal pricing strategy. For example, a respondent might prefer a huge umbrella in winter and a small and portable one in summer. Only considering one point in time in a conjoint experiment might lead the researcher to make false conclusions regarding consumer preferences. Dubé (2014) argues for heterogeneous consumers in the case of durable goods adoption with regard to individual discount rates. In other words, it is shown that consumers differ in how they discount future consumption utility. While succeeding to incorporate time effects, Dube (2014) does not account for respondent uncertainty. This can be considered a shortcoming because the passage of time influences many uncertainty related variables. Over time products/services might become cheaper, available or even obsolete and there is a loss in information if respondents cannot express this form of uncertainty.
2 This paper is relevant for two streams of marketing literature. Firstly, it is providing new insights to in the domain of conjoint based preference measurement. Secondly, it contributes to a substantial body of research that is concerned with multi-part pricing for subscription services.
Multi-part pricing has become a widely used practice in the market of service subscriptions at the start of this century (Lambrecht 2007). Nowadays, the practice of multi-part pricing occurs in various service industries. Examples include “cellular-phones, electricity distribution, vehicle leasing, retail banking and online retailing” (Gopalakrishnan 2014).
Furthermore, there seems to be consensus about the positive impact of multi-part pricing on firm profits. Bagh (2012) shows that three-part tariffs are a very efficient way to price discriminate in the presence of a large and heterogeneous market. Similarly, Lambrecht (2007) argues that subscription providers benefit from greater demand uncertainty by using three-part pricing. Nonetheless, there are also instances when multi-part pricing is not the optimal form of pricing. For instance, Goettler (2011) finds that multi-part pricing is not the optimal pricing scheme for an online grocer.
While showing the positive impact on firm profits Lambrecht (2007) also acknowledges the negative impact on the consumer surplus that is caused three-part pricing. However, Goettler (2011) claims that the consumer welfare decrease is due to the so called “flat-rate bias” and not due to the presence of multi-part pricing.
Being subject to the flat-rate bias means to prefer a flat-rate in a scenario where a pay per use arrangement would be the optimal choice from an economic perspective. In the literature it as further been argued that a flat-rate bias is caused by insurance, taxi-meter and overestimation effects (Lambrecht 2006). Simply put, having a flat-fee reduces the mental burden of choice making and, as can be seen from the popularity of all-inclusive vacations, this is valued by consumers.
The finding that the flat-rate bias is the main cause for the consumer welfare decrease is derived by including different tariff options in the choice design and comparing their impact on consumer welfare. In other words, the negative impact on consumer welfare depends on the availability of tariff options. If there are only multi-price subscriptions than consumer welfare should decrease as stated by Lambrecht (2007). If consumers can freely choose between multi-part pricing, a flat fee or pay per use, the strongest negative impact is caused by having a flat fee.
3 users seem to overpay on their flat fee and the few heavy users run up their bill by paying for their excess usage.
This paper is going to extend the research on multi-part pricing in the market of mobile subscription services because it marks the first time that a data-volume based multi-part pricing scenario is considered. It is also one of the first time that respondents are exposed to two attributes with consumption allowances (minutes allowance, data allowance). Furthermore, the inclusion of the time dimension provides insights on possible trends in customer preferences regarding different forms of pricing over time.
Overall, this research shows how time dynamic preference data could be elicited and how this additional information adds value to the measurement of consumer preferences. More specifically, the empirical findings suggest that it is possible to derive distinct preferences on the individual level for different moments in time. Additionally, it is found that contract-offer optimization and the evolution of segment heterogeneity are suitable fields for the utilization of this data. In terms of multi-part pricing a trend towards data volume based multi-part pricing and flat fee minutes pricing can be found.
This paper is organized as follows: In the first part, the used research framework and methodology are presented. Next , the most relevant sample statistics as well as estimation results are provided. After showing the extent to which preferences differ over time , two practical applications of time-dynamic preferences are presented.
1. Research framework
Conceptual model
4 In CBC respondents can choose between alternatives that differ in regard to their levels concerning the same set of attributes. The recorded choices are the dependent variable. The conceptual model depicted above has two features that are different to conventional CBC as applied by Iyenagar (2008). Difference number one is the use of choice probabilities over discrete choices and difference number two is the inclusion of the time dimension. Note that the use of choice probabilities simply means that the dependent variable is measured differently, while the underlying concept remains the same.
Time is included in the form of two sets of dependent variables. The first one represents the individual utilities as of today. The second one is the individual utility in one year from now. Note, that all choices are made today and that “time” actually refers to the moment at which the respondents would acquire a given choice. Therefore, it is easier to think of “time” as the time horizon of decision making. Time is measured as a discrete variable that can take the form of zero (now) and one (one year from now) depending on the choice scenario that is presented to the respondents.
Hypothesis 1 to 6 (table1) are about the immediate effect of the attributes/levels shown to the respondents. This kind of effects are the backbone of choice based conjoint and have been successfully derived in many applications over the years. Moreover, the hypothesized effects are based on findings from previous research ( Iyenagar 2008; Dippon 2011; Kusmanovic 2013).
Hypothesis 7-12 are about the mean differences between the two sets of individual utilities. Inference regarding these hypothesis allows for the assessment of trends over time. It is hypothesized that the preferences for data volume and optional excess data increase over time. The utilities for minutes allowance, automatic excess data, slower internet excess data and contract length are hypothesized to decrease over time. In general, a trend towards higher preference for data and lower preference for minutes is hypothesized. This is based on the observations that the amount of data volume has become main determinant for subscription prices. In other words, it is hypothesized that the trends that have shaped the telecommunication industry in recent years are to continue.
5 now. As argued above, testing for hypothesis 13 is improved due to the elicitation of choice probabilities rather than discrete choices.
Direct effect of attributes/levels on choice probabilities
Trends over time in individual preferences
Trend over time in individual preference uncertainty
H1: Data volume allowance has a positive effect on the subscription choice
H7: The positive effect of data volume allowance increases over time
H13: Time has a positive impact on the standard deviation of the stated choice probabilities
H2: Automatic payment per mb has a negative effect on the subscription choice
H8: The negative effect of “automatic payment per mb” increases over time
H3: Slower internet arrangement has a negative impact on the subscription choice
H9: The negative impact of slower internet increases over time
H4: The (Optional)-added data arrangement has a positive impact on the subscription choice
H10: The positive impact of optional-added data increases over time
H5: Minutes/sms allowance has a positive effect on the subscription choice
H11: The positive effect of minutes/sms decreases over time
H6: Contract length has a negative effect on the choice of subscription
H12: The negative effect of contract length on consumer choices is constant over time
Table 1 Hypothesis
1.2 Dependent variable
6
1.3 Control variables
Age, gender and income are included as the demographic control variables. Since the study focuses on university students, income is going to be referred to as available monthly budget.
Since customers might differ in how satisfied they are with their current contract, a net promotor score regarding their current subscription will be included as a control variable.
Furthermore, respondents are asked to state how much time is left on their current subscription. This variable is important because respondents who are at the end of their contract might be more engaged in the decision process. The same holds for respondents who recently subscribed to a new mobile contract.
1.4 Attributes
The second group of input variables are the subscription attributes which are given to the respondents in their choice task. Hereby, monthly fee and excess data arrangement represent price related variables. While the inclusion of the monthly fee is straight forward, “excess data arrangement” needs further explanation. These arrangements state what happens if a customer has used up his or her data volume. Predicting how much data volume to use can be at least as difficult as to predict how much minutes to use. Therefore, different “excess data arrangements” should be valued according to how certain or uncertain a customer is about their own data volume consumption. And since using excess data comes at a certain cost, it is reasonable to speak of a three-part pricing situation. This is why excess data derangement is included as an attribute in the choice design.
In contrast to Iyengar (2008) data volume allowance is included as an attribute to the expense of internet access ability. This is the case because smartphone diffusion is substantially higher in 2016 than it was in 2008. Furthermore, common practice shows that today’s offerings are differentiated on the basis of data volume allowance while internet ability is a given.
The variables data volume allowance and contract length are included because they are key attributes in the subscription decision. The importance of these two attributes becomes evident by observing the way subscriptions are presented in practice.
7 subscription. Including “phone” attributes does not provide additional information on the effect of multi-part pricing which is the purpose of this paper.
The service provider is not included as an attribute following Iyengar (2008) who find a weak brand effect.
1.5 Levels
For the price variable monthly fee three levels are used. More generally they can be interpreted as low medium and large1. The usage of three price levels follows the reasoning that it is better to use less more precisely measured price levels (Omre 2013) .
Three levels are used in the case of minutes and data allowances. Again the levels can be interpreted as small, medium and large2. For the minutes allowance, large is represented by the “unlimited minutes” level. This level is special because it represents a flat-fee contract in terms of minutes.
The excess data arrangement attribute can take on three levels. The first level is “limited internet speed” which implies a reduction in download and upload bandwidth. Automatic payment per mb is the second level and it represents a controversial practice where customers are able to continue to use the internet at full speed while being charged 1€ for every additional 100mb. The third possible arrangement is the option to buy additional data volume at the cost of 1€ per 100 mb. Note that level two and three only differ in the degree of automation by which extra charges occur.
In the case of contract length levels include 6, 12 and 24 months.
It is important to see that different levels actually can be interpreted as different business models. This is the case for excess data arrangement and minutes allowance. Respondents can potentially be given a choice between a flat priced subscription, a three-part priced subscription or even a five-part priced subscription. A flat priced subscription implies unlimited minutes with slower internet after the data volume is used up. One example of a three-part priced subscription is 1200 minutes with excess payment of 0.3€ per minute and slower internet after the data volume is used up. If respondent are presented with exhaustible minutes and a charge per additional mb used, five-part pricing occurs. The five parts are made up by (1) monthly fee, (2) minutes allowance, (3)data volume allowance, (4) excess minutes fee and (5) excess data fee. To simplify things, the remainder of the article will distinguish between data and minutes based multi-part pricing. Minute based multi-part pricing is represented by the 100 minutes and 1200 minutes allowance levels, while data-volume
1
Monthly fee: Low=5€; Medium=20€; Large= 40€
2
Minutes: Low=100 minutes with excess payment of 0.3€ per minute; Medium=1200 minutes with excess payment of 0.3€ per minute ; Large=unlimited minutes
8 based multi part-pricing is represented by the attributes data volume allowance and excess data arrangement.
1.6 Experimental Design
In order to gain insight on the effect of time on participant uncertainty two scenarios with different time horizons are created. In scenario one the participant is asked to imagine that his/her contract ends today and the presented choice set represents all available subscriptions that would start the next day.
Since telecommunication is regarded as important to students (Aljomaa 2016) and the sudden ending of the current contract is an unexpected event, respondents might experience time pressure as well as a prevention focus in their decision making. This form of decision making is described as a “emergency purchasing situation” (Samson 2014) and should trigger respondents to make clear cut decisions, which means stating high percentages in the context of choice probabilities.
The second scenario is included to enable a comparison of the parameters and their distributions. In order to evoke forward looking decision making respondents are told to imagine that they want to purchase a mobile subscription one year from now.
In both scenarios respondents are told that they can attach any available phone to their subscription. Note, that phone availability is different in both scenarios. This is the case because new phones might be released between the first scenario (today) and the second scenario (at the end of your contract). Furthermore, a forward looking consumer might foresee a price decline in the time span between the two scenarios. Phone availability and price developments represent resolvable uncertainty that is argued to be captured by using choice probabilities. To rule out question-order and fatigue effects half the respondents are first shown scenario one while the other half is exposed to scenario two first.
1.7 Choice set design
9
1.8 Factorial design
Based on the chosen attributes and levels, there are 243 possible subscriptions. The combination of 5€ and 10 GB as well as 40€ and 0 GB are prohibited to filter out unrealistic level combinations. This measurement reduces the number of possible subscriptions to 189.
2. Methodology
The estimation is done using the Sawtooth software and therefore the estimation procedure is in line with the Sawtooth Software Technical Paper Series (2013).
The utility of respondent i regarding subscription j is expressed using part worth utilities.
1. 𝑢𝑖𝑗= 𝛽𝑖𝑃𝑟𝑖𝑐𝑒𝑗+ 𝛽𝑖𝑀𝑖𝑛𝑢𝑡𝑒𝑠/𝑠𝑚𝑠𝑗+ 𝛽𝑖𝐷𝑎𝑡𝑎𝑗+ 𝛽𝑖𝐸𝑥𝑐𝑒𝑠𝑠 𝐷𝑎𝑡𝑎𝑗+ 𝛽𝑖𝐶𝑜𝑛𝑡𝑟𝑎𝑐𝑡 𝐿𝑒𝑛𝑔ℎ𝑡𝑗
There are two levels in Hierarchical Bayes estimation. At the higher level it is assumed that individual part worth’s can be described by a multivariate normal distribution. Including the error term on the right hand side shows that the random utility model is used. Note, that βi represents a vector of utilities of individual i and it is assumed to be normally distributed with mean b and variance W. 𝑿𝒊𝒋𝒕 represents the attributes of alternative j in choice set t.
2. 𝑈𝑖𝑗𝑡=𝛽𝑖′𝑋𝑖𝑗𝑡+ 𝜀𝑖𝑗𝑡 𝑤𝑖𝑡ℎ βi ∼ MVN(b, W)
The lower level assumption is that given a respondents’ parameters, choice probabilities are governed by a multinomial logit model (equation 4 and 5). It is important to keep in mind that 𝒚𝒊𝒋𝒕 is a value between 0 and 1, since it represents the stated choice probabilities.
3. L(𝑦𝑖 |β𝑖 ) = ∏ ∏ 𝑃𝑡 𝑗 𝑖𝑗𝑡𝑦𝑖𝑗𝑡 = ∏ 𝑃𝑡 𝑖1𝑡𝑦𝑖1𝑡𝑃𝑖2𝑡𝑦𝑖2𝑡𝑃𝑖3𝑡𝑦𝑖3𝑡
10 4. 𝑃𝑖𝑗𝑡=∑ exp(𝑈exp (𝑈𝑖𝑗𝑡) 𝑖𝑘𝑡) 5. 𝑃𝑖1𝑡= exp (𝛽𝑖 ′𝑋 𝑖1𝑡+𝜀𝑖1𝑡) exp(𝛽𝑖′𝑋𝑖1𝑡+𝜀𝑖1𝑡)+𝑒𝑥𝑝(𝛽𝑖′𝑋𝑖2𝑡+𝜀𝑖2𝑡)+𝑒𝑥𝑝(𝛽𝑖′𝑋𝑖3𝑡+𝜀𝑖3𝑡)
By taking the integral of equation 3 the “unconditional” likelihood is derived, which means that 𝑦𝑖 is unconditional on βi, while still being conditional on b and W.
6. L(𝑦𝑖 |b, W) = ∫ 𝐿(𝑦𝑖|β)Ф(𝛽𝑖|b, W)dβi
The prior of b is normal and the prior of W is an inverted Wishart distribution. In equation 6 Ф(𝛽𝑖|b, W)represents the normal density with mean b and variance W. The estimation is done using the Sawtooth software and therefore the estimation procedure is in line with the Sawtooth Software Technical Paper Series (2013).
The parameters of ßi b and W are estimated by an iterative process. This means that first ßi and W are used to estimate b. In the next step b is used to draw a new estimate of W. After that b and W are used to generate new estimates of ßi. Furthermore, this circle of estimation and re-estimation is continued for a long period of time. The likelihood of each respondents data is calculated conditional to the current state of that respondents part worth utility. This means there are two sources of information that must be balanced against each other.
3. Data
11 post-paid (47%) users raises some issues in terms of data preparation. Most importantly the survey produced some meaningless data, as even pre-paid users had to answer contract related questions. The problem was solved by creating a pre-paid category for the contract related questions. In the process of data preparation it was found that the contract related variables ( current minutes, current data volume, subscription type) are highly correlated. Therefore, it is decided to merge these variables into one variable called “contract value”. The merging is done by giving a monetary value to each contract characteristic. The valuation is based on average prices that are derived from the Dutch consumer society (consumentenbund.nl).
7. 𝐶𝑜𝑛𝑡𝑟𝑎𝑐𝑡 𝑣𝑎𝑙𝑢𝑒 = 𝑉𝑎𝑙𝑢𝑒 𝑑𝑎𝑡𝑎 + 𝑉𝑎𝑙𝑢𝑒 𝑚𝑖𝑛𝑢𝑡𝑒𝑠
For prepaid users Contract value is equal to zero . Therefore, the new variable contains information on the subscription type as well as the extend of the subscription. The average value of the current contracts is equal to 11.90€ with a standard deviation of 11.22€. It makes sense to compute the average contract value for post-paid users only, which is equal to 19.42€ with a standard deviation of 7.72€.
In total 8952 choice probabilities are elicited of which 4464 were part of the first scenario and 4488 of the second scenario. The difference in these two values is caused by one respondent who exclusively filled out responses for the second scenario (4*6=24). From the frequency distribution (appendix D) it can be seen that zero is used about 50% of the time. When looking at the non-zero responses it can be seen that 100% is used most often (778 times) followed by 10% (631 times) and 20% (601 times). Moreover, a clear tendency towards round numbers can be observed. For example if a respondent chooses 80% or 90% instead of 85% (figure 2).
VARIABLE CLASSIFICATION % MEAN MEDIAN STANDARD DEVIATION
12 NPS VALUE OF CONTRACT ELAPSED TIME 800-1500 1500-2500 >2500 Range from 0-10 Euro Seconds 28.3 2.7 2.7 7.28 11.9 550 1.91 11.22
Table 2 Sample demographics
From table 2 it can be seen that 53% of the respondents are male, while 46.6% are female. About two thirds of the respondents state to have a monthly budget of 0-800€ and 94% have an income below 1500€. In terms of the current level of customer satisfaction it is found that there are 21% detractors and 27% promotors. This results in a Net promotor score of 6 (27%-21%).
Figure 2 Frequency of choice probabilities
3.1 Estimation
The hierarchical Bayes model is estimated using Sawtooth software. Hereby, 10,000 iterations are made to find the set of individual utilities that have the highest likelihood given the choice data and a set of covariates. The covariates include the demographic variables (age, gender, income) as well as the subscription related information (NPS, Contract value). Furthermore, a variable indicating which scenario was shown first is added.
0 200 400 600 800 1000
13 MODEL PERCENTAGE CERTAINTY RLH AVERAGE VARIANCE PARAMETER RMS #PARAMETERS
CONJOINT1: PART WORTH 0,261 0,359 0,967 1,618 11
CONJOINT1: LINEAR 0,217 0,338 0,217 1,450 9
CONJOINT2: PART WORTH 0,215 0,337 0,82 1,369 11
CONJOINT2: LINEAR 0,178 0,320 0,806 1,202 9
Table 3 Estimation reports for both conjoint studies
For each conjoint exercise a part worth model and a linear model are estimated. For clarification, Conjoint 1 is referring to the “your contract ends today” scenario, while Conjoint 2 refers to the “your contracts end is one year from now “scenario.
The “goodness of fit” can be seen from the percentage certainty or the RLH values in table 3. In this regard it can be seen that all models are between 17% (conjoint1: part worth) and 26% (conjoint2: Linear) higher in terms of log likelihood than a chance model. From the Sawtooth estimation output in appendix B it can be seen that the number of iterations were sufficient as the “goodness of fit” statistics are stable over the last thousands of iterations (Orme 2013).
3.1.1 Part worth specification
Initially a part worth model is considered. To get an understanding of the estimation results the population averages of conjoint 1 and 2 are derived and depicted in figure 3.
Figure 3 Population averages for conjoint1 (contract ends now) and conjoint 2(contract ends one year from now=
The observed shapes of the average utilities are used to conclude a certain degree of face validity, as the signs of the parameters seem reasonable and intuitive. Furthermore, the linear pattern that can be seen for the monthly fee and data volume allowance lead to the decision to re-estimate the model with linear specification of these two attributes.
14
3.1.2 Linear specification
A model with linear specifications for the attributes price and data volume is estimated. To show that the resulting models are significantly better than a null model, a likelihood ratio test is conducted.
8. 𝐿𝐿(0) = #𝑟𝑒𝑠𝑝𝑜𝑛𝑑𝑒𝑛𝑡𝑠 ∗ #𝑐ℎ𝑜𝑖𝑐𝑒 𝑠𝑒𝑡𝑠 ∗ ln (14) = 185 ∗ 6 ∗ ln (14) = −1538.79
9. 𝐿𝐿(𝑙𝑖𝑛𝑒𝑎𝑟) = ∑1851 (ln (𝑅𝐿𝐻)𝑖∗ #𝑐ℎ𝑜𝑖𝑐𝑒 𝑠𝑒𝑡𝑠)
Equation 8 is used to derive the Log-likelihood of a null model, against which the linear model is tested. The log likelihood of the linear model is based on equation 9. From the output in table 4 it can be seen that both conjoint models fit the data better than the null model.
CONJOINT 1 CONJOINT 2 TOTAL LL -1176,37 -1230,28 LL(0) -1538,79 -1538,79 CHISQ. 724,83 617,01 DF = #PARAMETERS 9 9 CRITICAL VALUE 16,92 16,92 P(CHISQ) 0 0
Table 4 Likelihood ratio test
To check for potential interaction effects between the attributes/levels in the conjoint design, the Sawtooth interaction search tool is used. The tool includes all possible interactions and reports the gain in percentage certainty that would result when the interactions are included in the model. Hereby, all percentage gains are below 1% which suggests that no interaction effects are present (see appendix L) . This means that the inclusion of interaction effects would not improve the hierarchical Bayes model. Overall, it is shown that the model that specifies price and data volume as linear attributes is significantly better than the null model. Therefore, the inclusion of interaction effects is not necessary. Therefore, the “linear specification” model without interaction effects will be used to draw inference regarding the hypothesis testing.
3.2 The impact of the attributes and levels on today’s utility
15 most important attributes in the subscription choice are the monthly price (x̄=46%; σ =14%) and the data volume allowance (x̄=26%; σ =15%). The remaining attributes together account for about 30%.
Figure 4 The average attribute importance for conjoint 1
The incremental (average) willingness to pay is shown in figure 5. It can be seen that the willingness to pay for an additional GB of data is 2,68€ with a standard deviation of 3,26€. Furthermore, consumers are willing to pay almost 10€ to have unlimited minutes instead of only 100 minutes (σ=12,59€). Moving from automatic to optional excess data is valued by the consumers, with an willingness to pay of 4,50€ (σ=7,39€). Moreover, having a 6 month rather than a 24month contract contributes to a willingness to pay of 8.30€ (σ=17,28).
Figure 5 Incremental Willingness to pay for selected attributes/levels
The population averages of the linear model are used to test for H1 till H6. The sign of the average attribute/level utilities is used to draw inference of on the effect of these attributes on choices.
Positive parameters are found for data volume (x̄=0.15; σ=0.58), 1200 minutes (x̄=0.025; σ=0.53) ,unlimited minutes (x̄=0.5; σ=0.59), slower interne excess data (x̄=0.05; σ=1.24), optional excess data (x̄=0.19; σ=0.55) as well as 6 month contract length (x̄=0.45; σ=0.71).
Negative parameters can be observed for price (x̄=-0.77; σ=0.58) , 100 minutes (x̄=-0.53; σ=0.5), automatic excess data (x̄=-0.24; σ=1.01), 12 month (x̄=-0.06; σ=0.6) and 24 (x̄=-0.39; σ=0.98) month contract length.
Table 5 gives an overview of findings regarding the previously stated hypothesis. From the hypothesis that are about today preferences, only hypothesis 3 needs to be rejected, since slower internet does
0% 10% 20% 30% 40% 50%
Price_c1 Minutes_c1 Data_c1 Excess data_c1 Length_c1
€ - € 2,00 € 4,00 € 6,00 € 8,00 € 10,00 € 12,00 WTP € per Gig_c1
16 not have a negative impact on the subscription choice on average. It is worth mentioning that the high standard deviation shows that responded are heterogeneous in their preferences regarding the slower internet excess data arrangement.
HYPOTHESIS RESULT BASED ON OUTPUT
H1: DATA VOLUME ALLOWANCE HAS A POSITIVE
EFFECT ON THE SUBSCRIPTION CHOICE
Average utilities of conjoint1
H2: AUTOMATIC PAYMENT PER MB HAS A NEGATIVE
EFFECT ON THE SUBSCRIPTION CHOICE
Average utilities of conjoint1
H3: SLOWER INTERNET ARRANGEMENT HAS A
NEGATIVE IMPACT ON THE SUBSCRIPTION CHOICE
rejected Average utilities of conjoint1
H4 THE (OPTIONAL)-ADDED DATA ARRANGEMENT
HAS A POSITIVE IMPACT ON THE SUBSCRIPTION CHOICE
Average utilities of conjoint1
H5 MINUTES/SMS ALLOWANCE HAS A POSITIVE
EFFECT ON THE SUBSCRIPTION CHOICE
Average utilities of conjoint1
H6 CONTRACT LENGTH HAS A NEGATIVE EFFECT ON
THE CHOICE OF SUBSCRIPTION
Average utilities of conjoint1
H7: THE POSITIVE EFFECT OF DATA VOLUME
ALLOWANCE INCREASES OVER TIME
rejected paired sample t test on individual utilities from conjoint 1 and 2
H8: THE NEGATIVE EFFECT OF “AUTOMATIC
PAYMENT PER MB” INCREASES OVER TIME
rejected paired sample t test on individual utilities from conjoint 1 and 2
H9: THE NEGATIVE IMPACT OF SLOWER INTERNET
INCREASES OVER TIME
paired sample t test on individual utilities from conjoint 1 and 2
H10 THE POSITIVE IMPACT OF OPTIONAL-ADDED
DATA INCREASES OVER TIME
paired sample t test on individual utilities from conjoint 1 and 2
H11 THE POSITIVE EFFECT OF MINUTES/SMS
DECREASES OVER TIME
partly confirmed
paired sample t test on individual utilities from conjoint 1 and 2
H12 THE NEGATIVE EFFECT OF CONTRACT LENGTH
ON CONSUMER CHOICES IS CONSTANT OVER TIME
partly confirmed
paired sample t test on individual utilities from conjoint 1 and 2
H13: TIME HAS A POSITIVE IMPACT ON THE
STANDARD DEVIATION OF THE STATED CHOICE PROBABILITIES
rejected paired sample t test on individual standard deviations from conjoint 1 and 2
Table 5 An overview of the outcome of the hypothesis testing
17 by “slower internet”. Automatic excess data is the least preferred level of the excess data arrangement attribute.
3.2.1 The effect of time on individual utilities
To test for the difference between the means of the two sets of dependent variables a paired- sample t-test is conducted. The null hypothesis in this case states that the mean difference score of the individual level utilities is equal to zero. Rejection of the null hypothesis lead s to the result that there is a significant difference in the mean values of a given attribute/level utility over time. A p-value equal of 0.05 is used as the alpha level that leads to rejecting the null hypothesis. To account for differences in scaling between both conjoint studies mean centering is applied.
Table 6 gives an overview of the attribute/levels for which significant differences exist between the two scenarios. It can be seen that utilities for price (monthly fee), unlimited minutes, data volume, slower excess data, optional excess data, 6 month and 12 month contract length are significantly different.
There are cases for which utilities increase and cases were the utilities decrease over time. The utilities for price (monthly fee), optional excess data arrangement and 12 month contract length increase over time. The group of utilities that are decreasing over time consist of unlimited minutes, data volume, slower internet excess data and 6 month contract length. An overview of the mean values of the attribute/levels for both conjoint studies can be seen in appendix F.
PAIRED SAMPLES TEST
Paired Differences Mean Std. Deviation
df P-value
PAIR 1* Price_c1 - Price_c2 -0,0635 0,280544 184 0,002398
PAIR 2 100min_c1 - 100min_c2 -0,07295 0,560278 184 0,078221
PAIR 3 1200min_c1 - 1200min_c2 -0,06033 0,437258 184 0,062131
PAIR 4* unlimited_c1 - unlimited_c2 0,133285 0,645709 184 0,00553
PAIR 5* Data Volume_c1 - Data Volume_c2 0,020103 0,10534 184 0,010202
PAIR 6* Slower_c1 - Slower_c2 0,117749 0,742014 184 0,032193
PAIR 7 Automatic_c1 - Automatic_c2 0,028681 0,517892 184 0,452262
PAIR 8* Optional_c1 - Optional_c2 -0,14643 0,598162 184 0,001051
PAIR 9* 6month_c1 - 6month_c2 0,130458 0,479674 184 0,000285
PAIR 10* 12month_c1 - 12month_c2 -0,19054 0,554997 184 0.000
PAIR 11 24month_c1 - 24month_c2 0,060084 0,435398 184 0,062103
PAIR 12 NONE_c1 - NONE_c2 0,019525 1,99483 184 0,894237
18 The mean differences of the significant attribute/levels in table 6 are used to draw inference regarding H7-12. It can be seen that positive effect of data volume decreases over time, which leads to the rejection of hypothesis 7. There is no trend in the preference regarding automatic excess data, which leads to the rejection of hypothesis 8. Moreover, the positive effect of slower internet becomes negative over time, this leads to H9 being confirmed. Optional excess data has an increasing (positive) effect on the subscription choice., this leads to H10 being confirmed.
There is no trend in the levels 100min and 1200min, while the effect of unlimited minutes decreases over time. This leads to H11 being partly confirmed. H12 is also partly confirmed as the negative effect of 24month contract length is constant over time. The effect of 6month is decreasing over time, while the utility for 12 month increases over time.
3.2.2 The effect of time important marketing variables at the individual level
While significant differences in the utilities are a sign that the proposed method is able to capture additional information on the future expectations of respondents, it is the post-processing of these utilities where the most relevant managerial information is gained. Therefore, it is decided to investigate the changes in attribute importance and willingness to pay over time. The result of this investigation is that none of the attribute/importance’s and willingness to pay measures are significantly different between the two scenarios (see appendix G). This means that there is no trend over time regarding these measures in the population.
3.2.3 The effect of time on uncertainty
The individual standard deviation in attribute/level utility can be interpreted as a measure of respondent uncertainty. Once more a paired sample t-test is conducted to draw inference on how this measure is impacted by time. The output (appendix J) shows significant differences in the standard deviations for most of the attribute/levels. It can be seen that the standard deviation is significantly lower in conjoint number two for the following attributes/levels: Price, 100min, 1200min, optional excess data, 6 month contract length, 12 month contract length and the none option. There is not a single attribute/level for which the standard deviation increases over time.
19
4. Discussion
The findings regarding H1 and H6 represents the output of conventional CBC settings. Herby, all findings make intuitive sense and come to no surprise for industry experts. This can be interpreted as a sign that the derived estimates are valid in general terms.
It is found that time has an indirect effect on consumer preferences (H7-H12). Moreover, the difference in individual utilities are rather subtle in the case of mobile subscriptions, as average attribute importance and individual willingness to pay are constant over time. This raises the question of how to utilize the two sets of utilities, an issue that will be the topic of the remainder of this paper.
The most surprising result is that on average preference uncertainty decreases over time (H13). This contradicts the intuition that time increases uncertainty. A possible explanation for this could be that in the “your contract ends today” scenario decision making is influenced by time pressure and prevention focus (Samson 2014). This “emergency purchasing situation” adds additional aspects to the subscription choice, which are not present in the second choice scenario.
Future research in time dynamic preference measurement should try to avoid the creation of an “emergency purchasing situation” by stating the initial scenario in a less dramatic manner. For example the respondent could be told: “imagine your contract ends within one month…”.
4.1 Multi part pricing
Multi part pricing has been a feature of subscription offers for many years. The results from this research suggest that consumer preferences move away from minutes based, to data volume based multi part pricing.
There is a negative trend in the utility of both levels that represent minutes-based multipart pricing (100 minutes with 0.15 cents per additional minute, 1200 minutes with 0.15 cents per additional minute). Preferences for having unlimited minutes, which avoids being subjected to multi-part pricing, is constant over time. This is why, a moving away from minutes based multi-part pricing is concluded.
Multi part priced data volume is represented by two levels in this study. The more troublesome level (from a consumer protection perspective) is the automatic excess data arrangement, while the optional excess data arrangement provides a higher degree of control to the consumer.
20 even a negative trend regarding optional data (table 6). All in all, this leads to the conclusion that multi part priced data volume can be a source of revenue for telecommunication providers for some time to come.
This finding is further strengthened by the fact that the average standard deviation is constant for data volume allowance as well as automatic and slower excess data arrangement. The constant standard deviation shows that uncertainty in preferences are unresolvable for the chosen timeframe. Since at least some part of this uncertainty is due to consumption uncertainty, profits can be made as consumers are likely to over or underestimate their data consumption. Customers are and will remain uncertain about how much data to consume and the passage of time does not resolve this uncertainty.
In the case of minutes-based multi part pricing a trend towards lower uncertainty for the 100 minutes and 1200 minutes level can be observed (appendix J). Lower uncertainty regarding these attributes can be interpreted as a sign for decreasing minutes consumption uncertainty. A possible explanation could be that customers of both segments have gathered a good understanding of their own minutes consumption over the years. While this argument can explain a downward trend in consumption uncertainty, it does not explain why customers are more uncertain regarding their minutes consumption in the emergency purchase condition. The final explanation regarding this finding will have to left to future research. For now it is important to simply acknowledge that preference uncertainty decreases for levels that imply multi-part priced minutes.
Interestingly, it can be shown that consumers who already own a contract have higher preferences for optional excess data than pre-paid users (appendix K). This finding suggest that there might be some sort of learning curve regarding the importance of the optional excess data arrangement.
More knowledgeable consumers will try to avoid price discrimination, by preferring unlimited minutes and optional excess data. The difference between data and minute based multi part pricing is that consumers are more experienced with of the former. All in all, it can be concluded that there is a trend away from minute based multi part pricing to data volume based multi part pricing.
21
4.2 Optimal subscription offers at the individual level
The added value of having choice data at different moments in time becomes evident when considering which contract to offer to a current customer. An individual who is at the end of his contract should be offered a different contract than a person who still has one year remaining. To illustrate this, the optimal contracts that for all respondents at both moments in time are derived.
The optimal contracts are derived by finding contracts that represent a slightly higher utility than the individual benchmark utility. The idea being that the higher utility contract would be preferable to the individual consumer while being the cost minimizing outcome for the producer.
The individual benchmark is different for pre-and post-paid users. For post-paid users the benchmark utility is equal to the utility of the current contract. This is in line with Iyenagar (2008), who also uses respondents current contracts as their “status quo” option. Since the monthly fee paid by the responded is unknown, it is assumed that respondents pay according to the average market prices for the features of their contracts (see appendix H).The utility of the no-choice option is used as the benchmark for pre-paid users. This makes intuitive sense as the no choice options can be interpreted as deciding to stick with using pre-paid.
It is assumed that there is a linear relationship between the cost of an offer and its utility. In this line of reasoning, a slightly higher utility offer is “optimal” because a lower utility offer will not be chosen by the customer and a higher utility offer is less cost efficient. Obviously, This way of finding the optimal contract is a simplification of reality. In practice and when unit costs are known, a more sophisticated method for finding the optimal contract can and should be applied. An example for this is given by Iyenagar (2008), who uses actual variable costs from the US market. Nonetheless, the derived optimal offers are sufficient for illustrating the impact of the utilization of time dynamic preference data.
Following the procedure describe above optimal contracts are found for both moments in time. An example of two optimal subscription offers for a randomly chosen respondent can be seen in table 7.
In total 540 possible and realistic subscription offers are considered. Note, that the linear attributes allow for more levels being considered than were initially shown to the respondents3. It is possible to find optimal contracts for about 97% of the postpaid users and only 62% of the prepaid users. Furthermore, the number of unique contract offerings is rather high throughout the population (83%). A unique contract being one that is exclusively optimal for a single respondent.
3
22
Current contract
(benchmark)
Optimal offer now Optimal offer in one year
Price 18.50€ 30€ 15€
Minutes 0-200 minutes unlimited unlimited
Data 2-10 Gig 4 Gig 2 Gig
Excess data unknown automatic Slower
Contract length 0 month 6 month 24 month
Table 7 Optimal contracts for respondent 43
To further illustrate how these contracts are derived the two market scenarios are depicted in figure 6 and 7. Today’s utility functions of a randomly chosen respondent are depicted in figure 6. They are called “todays“ utility functions because they are based on the set of utilities form conjoint number one. In other words the graph shows todays preferences for the current contract, the optimal offer of today and the optimal offer form the future.
The point of intersection of the benchmark and the optimal offer can be used to determine the consideration willingness to pay. It can be seen that the respondent is willing to pay more for the optimal offer of today than for the optimal offer from the future. Therefore, it is profitable to offer the optimal contract of today instead of the optimal offer from the future to that respondent.
In figure 7 the same contracts are used but they are evaluated with the utilities form conjoint number 2. Now it can be seen only the optimal future offer is above the benchmark at a price of 15€. This is why it could be optimal to offer the optimal contract from the future to this respondent.
Figure 6 Utility function at different price levels (the market for today) for respondent 43
0 0,5 1 1,5 2 2,5 3 15 20 30 40 Uti lity Price in €
23
Figure 7 Utility function at different price levels (the market for today) for respondent 43
Overall, it can be said that figure 6 and 7 represent two different markets with different optimal contracts. Since customer data on remaining contract length is very likely to be available for companies like T-mobile or KPN, it could be feasible to determine the more appropriate market scenario for a given customer. Having data on preferences in both these scenarios could lead to better contract offers and hence higher profits.
The high number of contracts that are optimal for just a single respondent further illustrates the advantage of deriving individual level parameters. It explains the development of subscription personalization tools such as the “Stel Samen & Stel Bij sim only” by T-Mobile, which allows customers to choose freely between different attribute levels.
4.3 Segmentation
Following the words of wisdom that” if you are not thinking segments, you are not thinking” (Levitt) it is decided to use a segmentation example to illustrate the advantages of having preference data of two moments in time. The advantages will become evident as it is possible to detect changes in between-segment heterogeneity for subscription offers over time.
From observing the distribution of utilities of various demographics, it is found that the current type of subscription is best suited for the sake of segmentation. Therefore, two segments are formed. Segment one is represented by the group of pre-paid users, while segment two is made up by post-paid users. To test for segment heterogeneity an independent sample t-test is conducted. For Scenario number one it is found, that prepaid users have higher preferences for unlimited minutes as well as shorter contract length. Post-paid users attach significantly higher utility to data volume and having an optional excess data arrangement (see appendix K). Not only are pre-paid users significantly different in their preferences compared to post-paid users they also differ in terms of income and satisfaction scores. On average the pre-paid users have 140€ less in terms of monthly budget. The net promotor score of post-paid users is equal to 0.14 (28% Promotors-0.14 Detractors)
-2 -1,5 -1 -0,5 0 0,5 1 1,5 15 20 30 40 Uti lity Price in €
24 while it is equal to -0.08 (25%Promotors-33% Detractors) for pre-paid users. To summarize, pre-paid users have significantly less income while being less satisfied with their current subscription.
CONJOINT 1 (NOW) PRE-PAID MEAN POST PAID MEAN P VALUE
IMPORTANCE DATA 0,22 0,29 .00
IMPORTANCE CONTRACT LENGTH 0,12 0,08 .00
WTP PER GIG 1,97 3,12 .02
WTP FOR OPTIONAL INSTEAD OF
AUTOMATIC EXCESS DATA
2,84 5,54 .02
Table 8 Marketing variables at the segment level if new contract starts today (only significantly different variables are shown)
In the next step the relative attribute importance’s and willingness to pay variables are calculated, using segment average utilities for each moment in time. It is found that the segments differ significantly in terms of relative importance of data volume and contract length. This holds for both moments in time (table 8; table 9).
The average data volume importance for the post-paid segment is equal to 29% for both moments in time, while it is equal to 22% (conjoint 1) and 23% (conjoint 2) for pre-paid users. The importance of contract length, which also remains a point of differentiation between both segments over time, is greater for the pre-paid segment.
CONJOINT 2(FUTURE) PRE-PAID
MEAN
POST PAID MEAN
P-VALUE
IMPORTANCE DATA 0,23 0,29 .01
IMPORTANCE CONTRACT LENGTH 0,11 0,09 .04
Table 9 Significant segment marketing variables if new contract starts one year from now
25 So it is possible to detect changes in segment heterogeneity when using choice data form two moments in time. Moreover, knowledge of these developments should be used to draw managerial implications. In the case of the mobile subscription industry this means that pre-paid users and post-paid users become more similar in terms of willingness to pay for data volume and having optional excess data. The obvious implication would be to target the segments accordingly.
1. by offering data intensive premium-priced subscriptions to the post-paid segment today rather than in one year from now
2. by offering low cost subscriptions with short contract length to the pre-paid segment today
3. In one year from now increasing homogeneity between the two segments should be answered with offering more homogenous contracts
4.4 Limitations and directions for future research
There are some limitations that need to be considered. First of all, students are used as respondents. This limits the extent to which conclusions about the whole mobile subscription market can be made. That being said, students do have an appropriate knowledge of and interest in the mobile subscription market. Therefore, it can be said that it is justifiable to use a student sample in this case as long as overgeneralization is avoided.
While it was possible to observe preferences regarding minutes and data based multi-part pricing, not all forms of pricing arrangements could be represented in this research. An example is the emergence of post-paid subscriptions that allow the customer to change price-related attribute levels at the beginning of each month. Moreover, pay per usage subscriptions without monthly allowances are not included in this research.
Another limitation is the fact that preferences for today and one year from now are recorded in one moment in time. It can be argued that this paper measures expectations about future preferences rather than actual preferences in one year from now. Panel data on the actual development of preferences could be used to put the findings of this research into a broader context. Therefore, future research into the similarity between expected and actual preferences data is needed.
26
5. Conclusion
Time dynamic preference measurement has the potential to provide an worthwhile extension to CBC and preference measurement in general. This research includes the time dimension by considering preference data for two moments in time for the market of mobile subscriptions. After showing that preferences within individuals are indeed changing over time, the benefits of utilizing this information are discussed.
The discussion of the results clearly shows the added value of including the time dimension into consumer preference measurement. Two practical examples from the market of mobile phone subscriptions are presented to illustrate how information on time dynamics can used to potentially increase market share and profits.
On the individual level it could be shown that the incorporation of future preferences can be used to find optimal offers for different moments in time. Neglecting the development in preferences would have led to sup-optimal contract offerings and hence lower revenue. This form of offer optimization seem especially worthwhile for products for which the exact moment of replacement is known, as it is the case for current customers subscriptions.
Developments in segment homogeneity is the other promising topic of research for which time dynamic preference data is well suited. It could be shown that in the case of pre-paid and post-paid users some parts of heterogeneity are constant over time, while other vanish. Knowledge of such developments can be very valuable for creating forward looking segmentation and targeting strategies.
27
References
Aljomaa, S. S., Al.Qudah, M. F., Albursan, I. S., Bakhiet, S. F., & Abduljabbar, A. S. (2016).
Smartphone addiction among university students in the light of some variables.
Computers in Human Behavior.
Bagh, a., & Bhargava, H. K. (2012). How to Price Discriminate When Tariff Size Matters.
Marketing Science, 2399(December 2015), 1–16.
Blass, A. A., Lach, S., & Manski, C. F. (2010). Using Elicited choice probabilities to estimate
random utility Models: Preferences for electricity reliability. International Economic
Review, 51(2), 421-440.
Dippon, C. M. (2011). Consumer preferences for mobile phone service in the US: an
application of efficient design on conjoint analysis (Doctoral dissertation, Curtin
University, School of Economics and Finance).
Dubé, J.-P., Hitsch, G. J., & Jindal, P. (2014). The Joint identification of utility and discount
functions from stated choice data: An application to durable goods adoption.
Quantitative Marketing and Economics, 12(4), 331–377.
Goettler, R. L., & Clay, K. (2011). Tariff choice with consumer learning and switching costs.
Journal of Marketing Research, 48(4), 633-652.
Gopalakrishnan, A., Iyengar, R., & Meyer, R. J. (2014). Consumer Dynamic Usage Allocation
and Learning Under Multipart Tariffs. Marketing Science, 34(1), 116–133.
Iyengar, R., Jedidi, K., & Kohli, R. (2008). A Conjoint Approach to Multipart Pricing. Journal of
Marketing Research, 45(April), 195–210.
Iyengar, R., Jedidi, K., & Kohli, R. (2008). A Conjoint Approach to Multipart Pricing. Journal of
Marketing Research, 45(April), 195–210.
Lambrecht, A., Seim, K., & Skiera, B. (2007). Does uncertainty matter? Consumer behavior
under three-part tariffs. Marketing Science, 26(5), 698-710.
Lambrecht, A., Seim, K., Vilcassim, N., Cheema, A., Chen, Y., Crawford, G. S., … Sahin, O.
(2012). Price discrimination in service industries. Marketing Letters, 23(2), 423–438.
Kuzmanovic, M., Radosavljevic, M., & Vujosevic, M. (2013). Understanding student
preferences for postpaid mobile services using conjoint analysis. Acta Polytechnica
Hungarica, 10(1), 159-176.
28
Orme, B. K. (2013). The CBC System for Choice-Based Conjoint Analysis. Sawtooth Software
Technical Paper Series, 8(360), 1–27.
Samson, A., & Voyer, B. G. (2014). Emergency purchasing situations: Implications for
consumer decision-making. Journal of Economic Psychology, 44, 21-33.
Tripathi, S. N., & Siddiqui, M. H. (2010). An empirical investigation of customer preferences
in mobile services. Journal of Targeting, Measurement and Analysis for Marketing,
18(1), 49–63.
Appendices
A. Attributes used in previous studies on mobile subscription
Iyenagar (2008) Dippon (2011) Kusmanovic (2013) Schulze Boeing (2016)
Service provider Mobile phone
operator
Access fee Monthly charge Monthly charge
Minutes Minutes Minutes
Per minute rate Per minute rate (excess)
Internet access (yes,no)
Data allowance per month
Free internet (yes,no) Data volume allowance
Rollover Transfer
Data download speed
Length of contract Length of contract
Type of phone Excess data
arrangement
Support(technical)
Number choice
Promotions
Account balance check
29
B. Estimation report Sawtooth
Conjoint 1
30
C. Utility curves for the linear model
31
E. Average utilities for conjoint 1
32
F. Paired Sample t-test on utilities (output SPSS)
Paired Samples Statistics
Mean N
Std. Deviation
Std. Error Mean
Pair 1 Price (monthly fee)_c1 -0,8178 185 0,60063 0,044159 Price (monthly fee)_c2 -0,7543 185 0,621108 0,045665
Pair 2 100min_c1 -0,56683 185 0,512105 0,037651 100min_c2 -0,49388 185 0,639938 0,047049 Pair 3 1200min_c1 0,005861 185 0,552544 0,040624 1200min_c2 0,066195 185 0,464083 0,03412 Pair 4 unlimited_c1 0,491856 185 0,610029 0,04485 unlimited_c2 0,358571 185 0,571592 0,042024
Pair 5 Data Volume_c1 0,133754 185 0,166096 0,012212
Data Volume_c2 0,113651 185 0,175598 0,01291 Pair 6 Slower_c1 0,026358 185 1,289326 0,094793 Slower_c2 -0,09139 185 0,735584 0,054081 Pair 7 Automatic_c1 -0,2716 185 1,046944 0,076973 Automatic_c2 -0,30028 185 0,80325 0,059056 Pair 8 Optional_c1 0,176128 185 0,56501 0,04154 Optional_c2 0,322557 185 0,388022 0,028528 Pair 9 6month_c1 0,440768 185 0,732846 0,05388 6month_c2 0,31031 185 0,872888 0,064176 Pair 10 12month_c1 -0,08118 185 0,628732 0,046225 12month_c2 0,109361 185 0,444269 0,032663 Pair 11 24month_c1 -0,4287 185 1,012124 0,074413 24month_c2 -0,48878 185 0,994343 0,073106 Pair 12 NONE_c1 0,879446 185 2,229822 0,16394 NONE_c2 0,859921 185 1,831783 0,134675
Paired Samples Correlations
N Correlation Sig.
Pair 1
Price (monthly fee)_c1 & Price
(monthly fee)_c2 185 0,895076 4,27E-66
Pair 2 100mb_c1 & 100mb_c2 185 0,545994 9,15E-16 Pair 3 1200m_c1 & 1200m_c2 185 0,642452 6,41E-23 Pair 4 unlimited_c1 & unlimited_c2 185 0,404248 1,15E-08
Pair 5
Data Volume_c1 & Data
Volume_c2 185 0,811319 1,57E-44
33 Pair 7 Automatic_c1 & Automatic_c2 185 0,87584 8,39E-60
34
G. Paired Sample t-test on important marketing variables
Paired Samples Test Mean
difference Std. Deviation Sig. (2-tailed) Pair 1 Price_c1 - Price_c2 0,010766 0,123179 0,236056 Pair 2 Minutes_c1 - Minutes_c2 -0,00688 0,071554 0,192268 Pair 3 Data_c1 - Data_c2 -0,00582 0,105412 0,453631 Pair 4
Excess data_c1 - Excess data_c2 0,002917 0,06119 0,517573
Pair 5
Length_c1 - Length_c2 -0,00098 0,073172 0,855953
Pair 6
WTP € per Gig_c1 - WTP € per Gig_c2 0,163041 2,869158 0,440568
Pair 7 WTP 100mb-unlimited_c1 - WTP 100mb-unlimited_c2 0,339926 39,64355 0,907283 Pair 8 WTP optional_c1 - WTP automatic-optional_c2 -4,39146 37,65758 0,114424 Pair 9 WTP 24 to 6_c1 - WTP 24 to 6_c2 2,770854 25,43713 0,140157 Mean Std. Deviation Pair 1 Price_c1 0,460485 0,148285 Price_c2 0,449719 0,157192 Pair 2 Minutes_c1 0,103446 0,075917 Minutes_c2 0,110331 0,082128 Pair 3 Data_c1 0,263232 0,152116 Data_c2 0,269052 0,147986
Pair 4 Excess data_c1 0,077127 0,060499
Excess data_c2 0,07421 0,059769
Pair 5 Length_c1 0,09571 0,070475
Length_c2 0,096688 0,074756
35
H. Averages prices used to value current contracts
I. Population averages for Attribute Importance and incremental
WTP
Mean N Std. Deviation Price 46% 185 0,148285 Minutes 10% 185 0,075917 Data 26% 185 0,152116 Excess data 8% 185 0,060499 Length 10% 185 0,070475 WTP € per Gig € 2,68 185 3,263352WTP to move from 100mb to unlimited minutes € 9,69 185 12,59224 WTP to move from automatic to an optional excess data arrangement € 4,52 185 7,410175 WTP for having a contract length of 6 month rather than 24 month € 8,29 185 17,32764
0,00 € 2,00 € 4,00 € 6,00 € 8,00 € 10,00 € 12,00 € 0 200 400 600 800 1000
Average price per minute/sms
0,00 € 5,00 € 10,00 € 15,00 € 20,00 € 25,00 € 30,00 € 35,00 € 40,00 € 0 2000 4000 6000 8000 10000 12000
36
J. The effect of time on uncertainty
Paired Samples Correlations
N Correlation Sig.
Pair 1
Price (monthly fee)_c1 & Price
(monthly fee)_c2 185 0,787968 0
Pair 2 100mb_c1 & 100mb_c2 185 0,937795 0
Pair 3 1200m_c1 & 1200m_c2 185 0,850209 0
Paired Samples Statistics
Mean N
Std.
Deviation Std. Error Mean
Pair 1 Price (monthly fee)_c1 -0,7949 185 0,321494 0,023637 Price (monthly fee)_c2 -0,83389 185 0,392036 0,028823 Pair 2 100mb_c1 0,238134 185 0,969898 0,071308 100mb_c2 0,134325 185 0,885646 0,065114 Pair 3 1200m_c1 0,080551 185 0,642382 0,047229 1200m_c2 -0,01539 185 0,606123 0,044563 Pair 4 unlimited_c1 0,108622 185 0,668737 0,049167 unlimited_c2 0,083664 185 1,186492 0,087233
37 Pair 4 unlimited_c1 & unlimited_c2 185 0,949661 0
Pair 5
Data Volume_c1 & Data
Volume_c2 185 0,804406 0
Pair 6 Slower_c1 & Slower_c2 185 0,918589 0
Pair 7 Automatic_c1 & Automatic_c2 185 0,89151 0
Pair 8 Optional_c1 & Optional_c2 185 0,815767 0
Pair 9 6month_c1 & 6month_c2 185 0,950631 0
Pair 10 12month_c1 & 12month_c2 185 0,854958 0
Pair 11 24month_c1 & 24month_c2 185 0,919403 0
Pair 12 NONE_c1 & NONE_c2 185 0,658037 0
Paired Differences t df Sig. (2-tailed)
Mean Std. Deviation Std. Error Mean
Pair 1 Price (monthly fee)_c1 - Price (monthly fee)_c2 0,039 0,242 0,018 2,194 184 0,029 Pair 2 100mb_c1 - 100mb_c2 0,104 0,338 0,025 4,182 184 0,000 Pair 3 1200m_c1 - 1200m_c2 0,096 0,343 0,025 3,799 184 0,000 Pair 4 unlimited_c1 - unlimited_c2 0,025 0,590 0,043 0,575 184 0,566 Pair 5 Data Volume_c1 - Data Volume_c2 0,011 0,099 0,007 1,527 184 0,129 Pair 6 Slower_c1 - Slower_c2 0,006 0,429 0,032 0,204 184 0,838 Pair 7 Automatic_c1 - Automatic_c2 -0,018 0,473 0,035 -0,528 184 0,598 Pair 8 Optional_c1 - Optional_c2 0,090 0,360 0,026 3,413 184 0,001 Pair 9 6month_c1 - 6month_c2 0,134 0,332 0,024 5,504 184 0,000 Pair 10 12month_c1 - 12month_c2 0,046 0,293 0,022 2,131 184 0,034 Pair 11 24month_c1 - 24month_c2 0,053 0,392 0,029 1,836 184 0,068
38
K. Segmentation
INDEPENDET SAMPLE T TEST (SEGMENTATION)
P-value 1200M_C1 0,003494552 UNLIMITED_C1 0,015816456 DATA VOLUME_C1 0,005742826 OPTIONAL_C1 0,001030596 6MONTH_C1 0,000497756 24MONTH_C1 0,010483434 BUDGET 0,016214367 NPS_1 0,010044023
Means of significant different utilities l
Screen2 N Mean Std. Deviation 1200m_c1 Pre-paid 70 -0,11837 0,593695 Post-paid 115 0,117149 0,478739 unlimited_c1 Pre-paid 70 0,633524 0,592851 Post-paid 115 0,417881 0,578555 Data Volume_c1 Pre-paid 70 0,110423 0,157262 Post-paid 115 0,177463 0,158798 Optional_c1 Pre-paid 70 0,025473 0,588535 Post-paid 115 0,295297 0,49742 6month_c1 Pre-paid 70 0,6803 0,886654 Post-paid 115 0,309685 0,536141 24month_c1 Pre-paid 70 -0,62917 1,460392 Post-paid 115 -0,25007 0,458416 Budget Pre-paid 70 1,267606 0,476827 Post-paid 115 1,513043 0,765085 NPS_1 Pre-paid 70 6,816901 2,269738 Post-paid 115 7,556522 1,601487 t df Sig. (2-tailed) Data Importance_now -3,311813281 183 0,001117
Contract Length importance_now 3,567820969 183 0,00046
WTP € per Gig_now -2,336796049 183 0,020532
WTP automatic-optional_now -2,434962881 183 0,015853
