Telecommunications Policy 41 (2017) 227–241

Contents lists available at ScienceDirect

Telecommunications Policy
journal homepage: www.elsevier.com/locate/telpol

Estimating demand for fixed-line telecommunication bundles MARK

⁎
Maciej Sobolewski , Tomasz Kopczewski
University of Warsaw, Faculty of Economic Sciences, Długa 44/50 St., 00-241 Warsaw, Poland

A R T I C L E I N F O ABSTRACT

JEL classification: In this study, a stated preference discrete choice experiment is applied to model subscription
C25 choices between standardized packages of fixed-line telecommunications services in a sample of
D12 individuals in Poland. We provide willingness-to-pay (WTP) estimates for telephony, broadband
L96 and pay TV offered as stand-alone services and dual or triple-play bundles. The main goal of the
Keywords: paper is to investigate substitutability and complementarity between various component services
Bundling based on estimated valuations. Considerable preference heterogeneity towards fixed-line
Discrete choice experiment services is identified and explained by service profile and usage intensity.
Willingness-to-pay
Broadband is the highest valued among fixed-line services. Fixed-line broadband and pay TV
Fixed-line telephony
are super additive in valuations indicating complementarity. We obtain evidence of fixed to
Fixed-line broadband
Pay TV mobile substitution in voice and data, driven by unlimited mobile voice plans and data plans
Fixed-to-mobile substitution with higher transfer limit. On methodological grounds, choice-based WTP estimates are more
adequate measure of consumer reservation prices for telecommunications services than self-
reported valuations.

1. Introduction

Over recent years, bundling has become a common sales format on various markets, including airlines, banking, and in particular
telecommunications. The main rationale for integrating different services in a package is that it increases revenues via extracting
more surplus from heterogeneous consumers. Not surprisingly, dual, triple-play and quadruple-play bundles, already command for
70% of subscription revenues in the telecommunications industry.1 According to recent Eurobarometer survey, in 2015 exactly 50%
of EU households purchased telecommunication services in packages (European Commission, 2015).2 Fixed-line triple play bundles
(i.e. packages consisting of telephony, broadband Internet and pay TV) were adopted by 7% of households. Interestingly, this
proportion has dropped by four percentage points since the previous survey, after a period of continuous growth over several years.3
Reversion of triple-play adoption trend has been induced by emergence of unlimited mobile voice plans. It reflects an ongoing shift
in preferences from fixed-line to mobile technology in telecommunication services. Since 2014, adoption of bundles with fixed-line
telephony dropped sharply by 15% points, while adoption of bundles with mobile telephony increased by 16 points. Fixed-to-mobile
substitution (FMS), driven by continuous enhancements of mobile access technology, can be expected to spread beyond voice to
video and data services. Above changes already have a significant bearing on the demand for converged fixed-line offerings and

⁎
Corresponding author.
E-mail addresses: maciej.sobolewski@uw.edu.pl (M. Sobolewski), tkopczewski@wne.uw.edu.pl (T. Kopczewski).
1
Digital TV Research (2013).
2
Data from October 2015. This percentage has increased by 4 points since the previous survey from January 2014 (European Commission, 2014), but still shows
considerable variation among member states. In some countries, like France, Belgium and the Netherlands, bundle adoption hits 63–87%, while in Poland, Czech
Republic, Slovakia it is much smaller (32–39%) due to less developed fixed network infrastructure.
3
Quadruple-play bundle (fixed triple-play with mobile telephony) was adopted only by 5% of EU households in 2015. This package does not seem to be a game
changer in Europe, because of increasingly obsolete fixed-line telephony.

http://dx.doi.org/10.1016/j.telpol.2017.01.011
Received 30 June 2016; Received in revised form 31 January 2017; Accepted 31 January 2017
Available online 07 February 2017
0308-5961/ © 2017 Elsevier Ltd. All rights reserved.
M. Sobolewski, T. Kopczewski Telecommunications Policy 41 (2017) 227–241

increase pressure on traditional operators. However before those changes will become fully visible in market statistics, they can be
traced on individual level by looking at how usage and service profiles affect consumer utility from fixed-line services. We argue that
elicitation of preferences and analysis of their heterogeneity can serve as an early indicator of upcoming changes in consumer
behavior, providing operators and policy makers with better understanding of evolving playing ground in the telecommunications
industry.
The objective of our study is to estimate willingness-to-pay (WTP) for the three major fixed-line telecommunication services:
telephony, broadband and pay TV offered as stand-alone services or in different dual or triple-play bundles. By disclosing
substitutability and complementarity patterns across different services, we are able to assess how much of additional value is created
by integrating them into the package. We provide insights into the prospects of fixed network operators, by looking at how usage
profiles and intensity affect utility from fixed-line broadband and pay TV. Methodologically, we model subscription choices over
packages of fixed-line services with stated preference discrete choice experiment (DCE) on a sample of Polish respondents. Then,
from the estimates of utility function parameters, we derive money-metric WTP estimates for particular services and compare them
with self-reported valuations collected from respondents. Our interest in indirect elicitation methods is motivated by the evidence on
weaknesses of self-reported valuations. We narrow a gap in literature related to the lack of studies which apply stated preference
method to disclose patterns of substitutability and complementarity for bundles with more than two items. This case is particularly
relevant for telecommunications because bundles contain a larger number of component services then usually analyzed in other
applications.
Increasing the size of the package from two to three items rises a few challenging issues. First, the number of potential offering
rises from three to seven, which complicates experimental design. Second, demand for telecommunication bundles is affected by
numerous relationships between component services. For example, for many users, the so called over-the-top (OTT) online voice and
video services such Skype or YouTube are to some extent substitutes for fixed-line telephony and pay TV. Therefore, reservation
prices for the fixed telephony and pay TV will be conditional on whether a consumer has access to fixed-line broadband. On the other
hand, multiple-play bundles such as broadband and television package may create additional value to consumers due to
complementarity between components. Complementarity might stem from various sources like the technical integration, or
convenience from managing several services on a single account.4 Another issue which adds complexity to the elicitation of
subscribers preferences for telecommunication bundles is the impact of mobile technologies. Numerous studies evidence fixed-to-
mobile substitution in voice (Barth & Heimeshoff, 2012; Grzybowski, 2014) and more recently also in data services (Grzybowski,
Nitsche, Verboven, & Wiethaus, 2014; Srinuan, Srinuan, & Erik, 2014). Technological substitution between fixed-line and mobile
technologies is accounted for in the study by exploring how utilities from fixed-line voice, data and video services are impacted by
unlimited mobile voice tariffs and data allowances as well as over-the-top video services.
We obtain parameters of all seven distributions of willingness-to-pay. By checking for (non-) additivity of median reservation
prices we can infer whether integrated telecommunications services create additional value. The main result of the paper establishes
complementarity of fixed-line broadband and pay TV. The study finds out that fixed-line voice is already an obsolete service,
especially for subscribers with unlimited mobile voice tariffs, as implied by the negative mean utility in comparison to the baseline
utility level for mobile voice. The study offers a contribution to the ongoing research on FMS in data and video services. With respect
to fixed-line and mobile broadband our results indicate partial substitutability and partial independence. This is a transitional stage
between initial complementarity and full substitutability which will emerge with an advent of unlimited, high-speed mobile
broadband access. Finally, the study identifies considerable preference heterogeneity and explains its sources with a set of
respondent characteristics such as subscribed services, use intensity or income.
The rest of the paper is organized as follows. In Section 2 we provide a literature review on bundling. In Section 3 we describe
details of our empirical study. Section 4 sets up the modeling framework and Section 5 presents estimation results. Finally, Section 6
provides summary and conclusions.

2. Literature review

Bundling is defined as a sale of two or more products in a package at a discounted price (Stremersch & Tellis, 2002). Since the
seminal works of Stigler (1963) and Adams and Yellen (1976) economic literature recognizes the two main forms of bundling,5 and
research agenda focuses on conditions under which they generate more profits than separate sales (see Venkatesh & Mahajan, 2009
for an overview). Assessment of global gains from bundling is a core issue from managerial standpoint. Interestingly, only few
studies adopted this perspective.6 For example, Eckalbar (2010) suggests that bundling is rarely better enough to justify its
implementation unless monopolist encounters favorable combination of low unit costs and strong negative correlation of valuations.
His results explain why bundling might be applicable mainly to low marginal cost industries such as information goods or ICT

4
Stremersch and Tellis (2002) list many ways in which integration creates value added, such as compactness, seamless interaction, reduced risk, interconnectivity
and enhanced performance. Integrated ICT-based products and services, such as smart TV and video on demand (VOD) serve as good examples of those benefits.
5
Under pure bundling, a consumer can buy only a package while under mixed bundling he may choose to buy either a package or any component separately. There
can be also other forms of packaging, like partial mixed bundling.
6
Localized approach looks at the improvement of profits from introducing a package with an infinitesimal small discount on the sum of prices for all components.
McAfee, McMillan, and Whinston (1989) establish sufficient condition for mixed bundling to (locally) dominate separate sales for any atomless joint distribution of
reservation prices. Recently, Chen and Riordan (2013) utilized copula approach to generalize sufficient and necessary conditions of McAfee et al. (1989) for bundles of
a size larger than two.

228
M. Sobolewski, T. Kopczewski Telecommunications Policy 41 (2017) 227–241

services.
Pure bundling enables greater rent extraction if heterogeneity of reservation prices for a package is reduced in comparison to
separate sales. This is guaranteed under negative correlation between reservation prices, however Schmalensee (1984) has shown
that in some cases bundling might dominate separate sales, even if correlation is moderately positive. Mixed bundling is a
discrimination tool between different types of consumers by enabling a purchase of single components and package. If unit
production costs are substantial, mixed bundling offers additional advantage over pure bundling by avoiding implicit subsidization
of consumers who are mainly interested in one of the goods (Adams & Yellen, 1976). Likewise, there are supply-side arguments for
bundling. For example, if production costs are sub-additive, bundling is a good way to explore economies of scope. In the context of
telecommunication industry bundling has been proved to offer some additional, advantages to operators: such as churn reduction
(Prince & Greenstein, 2014; Yang, 2013) or leveraging position in the market through product tying or so-called predatory bundling
(Krämer, 2009).
Whether potential advantages of bundling will be realized depends on knowing consumer valuations for individual services and
their packages. In telecommunications the information on customers valuations is incomplete. Operator does not know the exact
reservation prices of each individual customer but can disclose distributions of reservation prices in population, via for example
consumer surveys. Yang and Ng (2010) analyze bundling of cellular handset with tariff plan in both complete and incomplete
information cases. They show algorithms for finding optimal handset discount in a bundle and conclude that in certain situations
pure bundling can outperform mixed bundling (that is operators should resign from offering tariff plans and handsets separately).
Wu, Hitt, Chen, and Anandalingam (2008) analyze customized bundling– when customers can form a bundle of certain size from a
possibly large set of products - under incomplete information. They show that this strategy outperforms pure bundling when
customers are heterogenous and yields profits very close to complete information case. Hence there is no much of additional value
from information on exact reservation prices.
Despite routine use of bundling in information industries, little attention has been paid in empirical research to various ways of
measuring consumer valuations for bundles. Jedidi, Jagpal, and Manchanda (2003) show that bundling prices constructed from self-
reported reservation prices yield suboptimal profits compared to indirectly elicited valuations. Outside telecommunications, several
studies confirmed that significant differences exist between outcomes of indirect and direct methods for willingness-to-pay
elicitation (Backhaus, Wilken, Voeth, & Sichtmann, 2005; Wertenbroch & Skiera, 2002). The latter method is straightforward in
implementation but potentially exposed to a large measurement bias caused by embedding, reporting and scope effects. The
magnitude of that bias is product-specific, but it is relatively larger for higher-priced, less often purchased products such as
telecommunication services (Miller, Hofstetter, Krohmer, & Zhang, 2011). Direct elicitation based on self-reported point
declarations are also subject to hypothetical and strategic response biases, especially when consumers are uncertain about their
preferences. In such case combining range approach with incentive compatible conditions might produce unbiased results (Wang,
Venkatesh, & Chatterjee, 2007). Recently Carare, McGovern, Noriega, and Schwarz (2015) estimated self-reported valuations for
broadband by non-adopting households in the US. They find significant over-reporting of high values as well as under-reporting of
lower values and disclose socio-economic determinants of this bias.
Empirical contributions related to bundling in telecommunications focus mainly on factors determining adoption of packages by
consumers. Srinuan et al. (2014) find that apart from the size of a price discount, also socio-demographic profiles of subscribers
influence the probability of buying multiple services. The contribution from Üner, Güven, and Cavusgil (2015) finds large preference
heterogeneity for different telecommunications packages among Turkish subscribers. They report that young and well-educated
people are keen on adopting packages, perhaps due to a balanced structure of reservation prices which enables greater surplus
savings.
Literature related to the measurement of consumers’ willingness-to-pay for telecommunication services is scarce. Existing
studies have been mainly concerned with valuations of numerous technical attributes. Ben-Akiva and Gershenfeld (1998) is the first
study to apply stated preference discrete choice to model preferences over bundles composed of fixed-line telephony service and 13
call-enhancing add-ons such as call waiting, call forwarding or number presentation.7 They obtain utility parameters of price, single
add-ons and their combinations having a common functionality. Based on those estimates they are able to calculate willingness to
pay and assess complementarity and substitutability between different features. Klein and Jakopin (2014) estimate willingness-to-
pay for bundles composed of mobile voice and data in a conjoint study. They focus on five attributes of mobile plans: free minutes,
type of cellular handset, data transfer speed and limits, text messaging and price.8 Based on part-worth utilities willingness-to-pay
estimates are calculated. They conclude that features such as text messaging or subsidized smartphone have lower relative value than
unlimited mobile voice tariff or large monthly data limit. Recently, Grzybowski and Liang (2015) applied discrete choice modeling to
revealed preference data on mobile subscriptions bundled with ADSL or FTTH broadband in France. They show that WTP for
quadruple-play bundles is higher for FTTH than ADSL and dependent on several characteristics of mobile tariffs, such as unlimited
calls, phone subsidies or contract duration. Authors infer complementarity and substitutability relations between various
components of quadruple-play package.
Our study differs from the papers listed above in that we are primarily concerned with substitutability and complementarity

7
Due to large number of calling features they resign from classical attribute based approach towards constructing choice set and instead use extended menu
approach. Individual prices and bundle discounts had three levels and the choice set consisted of 54 menus obtained from fractional factorial design. Each respondent
made six choices between 3 and 6 preselected discounted packages and custom offer which he could construct from particular menu of all available features.
8
The number of attribute levels varied from 2 to 4 resulting in 384 possible alternatives, too much for implementation. Respondents had to rank order 16 bundles
constituting fractional factorial experimental plan.

229
M. Sobolewski, T. Kopczewski Telecommunications Policy 41 (2017) 227–241

relations between fixed-line services. Hence, we define preformatted services and do not analyze the impact of technical parameters
or enhanced add-ons on subscribers’ choice. In this respect our framework is similar as in Jedidi et al. (2003) who infer
substitutability or complementarity from (non)additivity of stated-preference valuations of two-component bundles of standardized
commodities. To accommodate for choice based willingness-to-pay for bundles with three component services we use reduced
experimental plan. However unlike previous studies we utilize efficient non-orthogonal design which allows for obtaining more
informative choices than orthogonal design while keeping the cognitive burden of choice experiment on acceptable level. In the next
step we add insights into FMS research by exploring sources of preference heterogeneity with usage profiles. This enables us to
capture the impact of both mobile and over-the-top services on the demand for fixed-line bundles.

3. Empirical study

Researchers can use either revealed or stated preference data to model consumer behavior. The former type of data represent
actual choices made under real budget constraint and is therefore considered less biased. However due to confidentiality reasons
revealed preference data is rarely accessible. Therefore researchers usually relay on hypothetical choices declared by individuals in
carefully designed discrete choice experiments (DCE). Importantly, results of preference modeling based on revealed and stated
preference data are highly consistent, provided that hypothetical choices meet incentive compatibility conditions (Carson, Flores,
Martin, & Wright, 1996; Vossler & Watson, 2013). DCE ensures sufficient data variability for model estimation and hence allows
for systematic study of most important factors that influence choice behavior. The data collected in choice experiment allows to
estimate the parameters of consumers’ utility function and then simulate their market behavior, calculate welfare changes and
estimate respondents’ willingness-to-pay for various levels of non-pecuniary attributes (Ben-Akiva & Lerman, 1985; Louviere,
Hensher, & Swait, 2006; Train, 2009).
To elicit consumer preferences for fixed-line bundles, a dedicated discrete choice experiment was designed and conducted on a
sample of individuals in Poland. Based on stated preference data we estimate the utility function which represents declared choices
and calculate willingness-to-pay for single services as well as dual and triple-play bundles. Finally we infer about substitutability or
complementarity of bundle components by checking additivity of obtained valuations, and explain preference heterogeneity with
usage intensity and service characteristics.

3.1. Choice experiment questionnaire

A dedicated questionnaire has been developed and tested for our research. It contained a block dedicated to the use of
telecommunication services by our respondents. We have asked about details of actually subscribed services, like volumes of free
minutes, TV channels, transfer speeds and monthly data allowances. We have also collected information on providers, adoption of
bundling offers and monthly costs of services.
In the part of the questionnaire dedicated to the choice experiment, respondents were introduced to the hypothetical situation of
moving into a new apartment in another town. They were instructed that their current fixed-line services are not geographically
portable. Hence, they need to choose a new package from an offer presented by a telco provider operating exclusively in the area. As
noted earlier, all three component services (TEL, TV, INT) had predefined technical parameters listed in the questionnaire.9 Those
parameters corresponded to actual ‘average’ market conditions and were invariant over respondents and choice situations. We
deliberately put aside technical attributes to focus entirely on the relations between different fixed-line services combined in
packages. Consequently our choice alternatives had only two attributes:

(1) type of fixed-line package,

(2) the monthly subscription fee for that package.

The ‘type’ attribute represents seven combinations of packages that can be formed out of three component services: three single-
component offers, three dual-play and one triple-play bundle. Subscription fee for each package, denoted in polish zloty (PLN), is
measured on a continuous scale within range 10–100. Package fees could differ considerably from respective market values to avoid
dominated alternatives in the choice situations. Both choice attributes are summarized in Table 1.
We had to address two problems related to designing a choice set for bundles of size three. With two items in the bundle it is
possible to apply full main effects orthogonal plan, as in Jedidi et al. (2003). Their plan has 27 choice situations which can be
conveniently split into three blocks. However, with three items, the ‘type’ attribute has seven levels. Consequently, even for a
minimal number of price levels, the full factorial plan has 2187 combinations – too many for implementation in any survey. As noted
in the literature review, this issue is usually resolved by applying fractional orthogonal design, where only some combinations of
attribute levels are used in a balanced pattern. Another problem was related to the fact that we could not present full set of seven
packages a single choice situation. As discussed in Louviere et al. (2006) choice tasks with more than four alternatives might
undermine reliability of collected data by imposing too heavy cognitive burden on respondents. To accommodate for both issues we

9
For example, broadband service was specified with 20Mbit/5Mbit downlink/uplink speeds, no transfer limit, a home router with Wi-Fi. Telephony plan had 5h of
free calls to landlines and mobiles and 0.20 polish zloty (0.05 euro) per minute price for additional calls (1EUR≈4PLN). Pay TV offer contained a detailed lists of
channels presented in thematic groups: films, information, sports, science, children. All of them were available in high definition format. Set-top-box had recording
functionality on 1GB HDD. All services had 12 months commitment period.

230
M. Sobolewski, T. Kopczewski Telecommunications Policy 41 (2017) 227–241

Table 1
The list of attributes and attribute levels used to describe choice alternatives.

Attribute Attribute level Definition

Type of fixed-line package [TV] Only pay TV

[TEL] Only telephony
[INT] Only broadband
[TV+INT] Bundle with pay TV and broadband
[TV+TEL] Bundle with pay TV and telephony
[INT+TEL] Bundle with broadband and telephony
[TV+TEL+INT] Bundle with pay TV, telephony and broadband
Monthly subscription fee for Fixed-line package (in PLN) [10,100] Any number between 10 and 100 on a continuous scale for fixed-line packages
Outside option [Mobile only] Denotes an outside option

applied efficient design. This approach minimizes standard errors of the utility parameters based on some prior information about
parameter values (Sándor & Wedel, 2001).10 Recent research has shown that efficient plans extract more information from
respondent choices than orthogonal plans (Street & Burgess, 2007). In our study, experimental plan was optimized with respect to
C-Error which minimizes a sum of variances for the ratio of two parameters. C-Error is a recommended optimization criterion for
problems involving calculation of willingness-to-pay, see Scarpa and Rose (2008).
Final experimental design had 36 single choice situations split into three blocks.11 Respondents were randomly assigned to one
of the blocks containing 12 choice tasks, each with four alternatives: three variants of fixed-line packages with corresponding prices
and the ‘no choice’ alternative. Respondents were instructed that (i) in a given choice situation they can choose only from the
presented subset of packages or choose not to buy; (ii) picking up ‘not to buy’ option implies, that in the new apartment they will use
only currently subscribed mobile services and (iii) buying one of the fixed-line alternatives does not preclude them from using
currently subscribed mobile services. These restrictions ensured that the framing of our choice experiment resembles real-life
situation, where respondents choose the best option from different combinations of fixed-line and mobile services. In particular,
respondent could decide not to buy any fixed-line service if she felt that her current mobile package is sufficient. An example of a
choice card shown to respondents is given in Fig. 1.
Together with declared choices, respondents provided us with their subjective valuations of particular services. Hence, we can
compare valuations provided directly in the survey with those obtained indirectly from a choice experiment.

3.2. Sample data

We have collected data from 272 respondents in an internet survey administered on a platform developed at University of
Warsaw. Respondents were incentivized for submitting a complete set of answers with a promised contribution to a charity.12
Although the collected sample is not representative of users of telecommunication services in Poland, we use it for the illustration of
how stated preference elicitation and modeling techniques can be employed for the analysis of bundling with three components.13
Still, some of our findings remain valid irrespective of the representativeness of the sample. In particular, this refers to the qualitative
impact of choice invariant characteristics, such as possessed services and their usage, on the utility from fixed-line bundles.
Descriptive statistics of the sample, shown in Table 2, allows to draw a couple of interesting observations. Looking at adoption
rates of particular services (part A) we observe that mobile telephony and fixed-line broadband are the two most popular
telecommunications services (with penetration of 99% and 85% respectively), followed by pay TV (59%) and fixed telephony (39%).
Fixed-line broadband remains a prevailing type of Internet access, yet 11% respondents already access the Internet from home,
using a 3G/4G modem with a dedicated SIM module. Cable remains a dominant technology for fixed-line access, which is specific for
Poland and other Central European countries due to relatively underdeveloped copper infrastructure. Broadband transfer speeds,
shown in part C, look quite satisfactory both for fixed and mobile type of access. Only 18% of our sample respondents (excluding
those unaware) have transfer speeds less than 10 Mb/s. An important difference between fixed and mobile access to the Internet is
related to monthly data allowances (part D). While fixed-line broadband is usually unlimited, mobile data plans have tight limits:
65% of respondents reported transfer allowance of up to 2 GB per month while only 3% used unlimited offers. For this reason, we
expect that the valuations for fixed-line Internet will still be substantial given that mobile access, albeit widespread does not allow for
comfortable video streaming.

10
We obtained priors from declared reservation prices collected in pretests of the questionnaire. The final design is presented in the Appendix.
11
We have resigned from using complete or partial ranking format due to problems with stability and reliability of such stronger orderings, see Louviere et al.
(2006) for discussion.
12
Respondents could choose from a preselected list of public benefit organizations and charity trusts representing a wide range of activities, like environment
conservation, medical and educational support for children or disabled persons. Data collection took one week and ended in May 2015. The average duration of an
interview in the whole sample was 16min (with a standard deviation of 5.5min) and complied with the expected duration time indicated in the instruction to the
questionnaire (15–20min).
13
Respondents from our sample are on average younger, better educated and more familiar with using Internet, particularly because they participated in a web-
based data collection method. These differences likely impact the estimation results for particular bundles: overstating the population valuations for broadband and
understating valuations for fixed-line telephony.

231
M. Sobolewski, T. Kopczewski Telecommunications Policy 41 (2017) 227–241

You are presented with the following offers of fixed services for your new apartment. Which of the following
package would you consider the best for yourself?

Offer A B C D

Type of package TV+Internet TV+Telephony TV None. I will

use only my
current mobile
Monthly subscri fee (PLN) 95 41 31 services

Your choice □ □ □ □

Fig. 1. Example of a choice card (translation).

Table 2
Descriptive statistics of sample respondents.

A. Adoption of services % N
[1] pay TV [TV] 59% 161
[2] Fixed-line BB Internet [INT] 85% 231 F. Demographics
out of which: cable 68% 158 Age (min/max/mean/median) 21/84/32/28
out of which: xDSL 32% 73 Women/men [%] 73%/27%
[3] Fixed-line telphony [TEL] 39% 106
[4] Mobile telphony 99% 270
out of which: voice+data plan 64% 174
[5] mobile internet for home use (extra sim) 11% 29

B. Adoption of bundles
Monthly expenses [PLN] Bundle discount
% N min max mean median
only [TV] 13% 36 10 130 52.97 50 –
only [TEL] 3% 8 10 119 42.53 37 –
only [INT] 31% 85 15 89 52.21 50 –
[TV+INT] dual play 19% 51 20 188 92.98 86 13%
[TV+TEL] dual play 1% 3 60 65 63.33 65 51%
[INT+TEL] dual play 9% 24 35 150 78.71 80 20%
[TV+TEL+INT] triple play 26% 71 30 200 112.4 110 31%

C. Transfer speeds at home

N 0–2 Mb/s 2–5 Mb/s 5–10 Mb/s 10–20 Mb/s 20–50 Mb/s More than 50 Do not know
fixed-line BB Internet 231 1% 4% 13% 21% 15% 19% 27%
mobile internet for home use 33 6% 6% 6% 27% 15% 3% 36%
(dedicated sim)

D. Mobile data allowance

N 1 GB 2 GB 3 GB 4–10 GB More than 10 GB Unlimited Do not know
monthly data limit for mobile Internet 174 37% 28% 9% 10% 1% 3% 12%

E. Monthly expenses on mobile phone

N 0–20 PLN 20–40 40–60 60–80 80–100 More than 100 Do not know
[pln] 270 6% 6% 6% 27% 15% 3% 36%

Second, telephony is the least popular among fixed services in our sample, which is expected, given both the ongoing trend of
fixed-to-mobile substitution and overrepresentation of ‘digital natives’ in the sample. Interestingly, fixed telephony is bought almost
entirely as a component of a dual-play bundle with internet (TEL+INT) or triple-play package with the internet and pay TV (TV+TEL
+INT). Another dual-play combination (TEL+TV) is unpopular despite the largest discount offered by the operators. Third
observation refers to the frequency of bundling adoption, shown in part B. Around 55% of our sample respondents use some form of
bundling, which exceeds the figure for Poland from already mentioned Eurobarometer survey. Triple-play and dual-play with TV
+INT are two the most popular bundles, adopted respectively by 26% and 19% of respondents. Pricing data reveal a clear discount
pattern, according to which more popular packages are priced with much smaller discounts than the less popular ones.

4. Econometric framework

In this section, we describe how discrete choice experiment data can be used to model respondents’ utility function. This function
enables to quantify the extent to which each attribute influences choices and determine on what terms they are willing to trade one
attribute for another. We discuss also how to calculate respondents' willingness-to-pay - a rate at which they are willing to exchange

232
M. Sobolewski, T. Kopczewski Telecommunications Policy 41 (2017) 227–241

their money for the different packages of fixed telecommunication services.

4.1. Econometric model

Formally, discrete choice modeling is based on the random utility model (McFadden, 1974). In this framework, the utility
function of consumer i ∈ I from alternative j ∈ J in choice situation t ∈ T can be expressed as:
Uijt = β′xijt + εijt (1)
where β is the vector of utility parameters, x is the vector of observed attributes specific to the consumer the alternative j and choice
situation t , and ε is the random component, representing the joint influence of all unobserved factors that influence decision-making.
By assuming that the random component is identically and independently Gumbel distributed, the multinomial logit (MNL) model is
obtained with familiar, closed-form expression for the choice probabilities of each alternative (Greene, 2011). In this study, we apply
a mixed logit (MXL) extension to take the respondents’ preference heterogeneity into account (Greene & Hensher, 2007). MXL
model allows that consumer i has specific parameters of the utility function: βi . Those parameters are non-observable but follow
distributions a priori specified by the modeler: βi~f (b, Σ), where b is the vector of the mean values of parameters and Σ is their
variance-covariance matrix (possibly non-diagonal to account for correlations across alternatives or choice situations). By assuming
a structured variation of individual tastes in the sample, in the form of individual-based parameters, the MXL model is more realistic
and typically yields a much better fit to the data, including also applications to telecommunications (Czajkowski & Sobolewski,
2016). This benefit comes at the cost of a more complicated estimation procedure. In a discrete choice experiment, Pijt – the
unconditional mixed logit probability of choosing alternative j in situation t by consumer i - is an integral of standard logit
probabilities over a density individual utility parameters Since mixed logit probabilities involve integrals which do not have closed
forms, unconditional probabilities must be simulated by taking multiple random draws from respective joint distribution and
averaging (Train, 2009). In the final step, the sequence of T choices made by each person during the experiment is represented by the
log-likelihood function from which estimators of b, Σ can be obtained numerically from maximization of the following log-likelihood
function:
I 1 D T J exp (x β )
ijt i
LL = ∑i =1 log ∑d =1 ∏t =1 ∑ j =1 yijt J
D ∑ j =1 exp (x ijtβi) (2)
where yijt is a dummy variable equal to 1 if respondent i selected alternative j in choice situation t and 0 otherwise.D represents the
number of draws taken from joint normal distribution.14
Initially, we assume homogenous means and variances of all K utility parameter across individuals. This implies that
βi = b + Σvi , where vi is individual specific unobserved heterogeneity with mean vector 0 and covariance matrix I and
Σ=diag(σ1,…,σk ) is the vector of standard deviations of distributions of βki across their means bk . Later on we explore preference
heterogeneity with the extension of MXL model, which allows distributions of random parameters to be heterogeneous with
observed respondent characteristics z i such as usage profile. Formally, we allow means of parameters to vary with observed vector of
characteristics zi while keeping constant covariance matrix: βi~f (b + Δzi , Σ), where Δ estimable vectors of parameters that enter
heterogeneous means of random parameters (Greene, 2011).15

4.2. Willingness-to-pay

With linear utility function, a consumer's willingness-to-pay for a change in an attribute k ∈ K is defined as the ratio between the
parameter of interest and the minus price attribute, as income is usually missing (Bliemer & Rose, 2013):
βk
WTPk =
βprice (3)
This is equivalent to calculating a marginal rate substitution between attribute k and monetary variable. In MNL model, both
coefficients are fixed, but uncertain due to a sampling variance. Hence, WTP given in Eq. (2) is, in fact, a random variable, for which
point estimate calculated from MNL coefficients might have distribution with undefined moments. This is the case, if βk , βprice are
normally distributed, making it difficult to obtain statistical properties of WTP. To overcome this problem WTP measure and
corresponding confidence intervals are calculated from a simulation (Krinsky & Robb, 1986). In MXL, the simulation of WTP has to
account that both coefficients are random variables following specific distributions assumed by the modeler. In this study we use an
extended version of Krinsky and Robb method in which instead of fixed coefficients, individual parameters from their assumed
distributions are drawn in a simulation (Bliemer & Rose, 2013; Hensher & Greene, 2003). More specifically, let
Y=(βk_μ , βk_σ , β , βprice_σ ) be the vector of estimated MXL parameters describing means and standard deviations of a random
price_μ
attribute k and price, both normally distributed. In the first step we generate a single random draw of each element of Y according to
the process given by:

14
The mixed logit model was estimated using Nlogit with 1000 Halton draws.
15
Individual variances can also be expected to differ with respect to selected characteristics of respondents, such as currently used services. However as this issue
falls outside the main topic we do not analyze models with heterogeneous variance here.

233
M. Sobolewski, T. Kopczewski Telecommunications Policy 41 (2017) 227–241

Table 3
Definition of variables used to model respondents’ choice of bundled offers.

Variable name and definition Measurement

TEL – fixed telephony standalone Dummy: 1=TEL; 0=otherwise

TV – pay TV standalone Dummy: 1=TV; 0=otherwise
INT – fixed Internet standalone Dummy: 1=INT; 0=otherwise
TELTV – telephony+TV dual-play Dummy: 1=TEL+TV; 0=otherwise
TELINT – telephony+internet dual-play Dummy: 1=TEL+INT; 0=otherwise
TVINT – tv+internet dual-play Dummy: 1=TV+INT; 0=otherwise
TELTVINT – telephony+tv+internet triple-play Dummy: 1=TEL+TV+INT; 0=otherwise
ONLYMOB – only mobile services at home Dummy: 1=if no choice alternative was selected;
0=otherwise
PRICE – monthly subscription fee Continuous from range [10,100]

Y͠ = Y + c′u (4)
where matrix c’ is Cholesky factor of the variance-covariance matrix of Y and u is the vector of standard normal variates. Then, for a
∼ ∼ ∼ ∼ ∼
single realization of Y =(βk_μ , βk_σ , β , βprice_σ ), we take S draws of individual parameters from respective distributions and
price_μ
calculate individual WTP values for attribute k according to the Eq. (5):
∼ ∼
βk _μ + βk _σ ∙v
ki, k =
WTP ∼ ∼
βprice _μ + βprice _σ ∙v (5)
where v is a multivariate standard normal distribution. In this way, information from entire distribution of individual utility
parameters is taken into account. The central tendency of individual WTP distribution is often characterized by median to avoid
exploding means (Hole, 2007). Next, by taking R draws of Y͠ , a distribution of median WTP with corresponding confidence intervals
can be obtained. We use the mean of median WTP distribution, as a final measure of willingness-to-pay in our sample.

5. Estimation results

The final dataset consisted of 3264 choices made by 272 respondents. We use these data to estimate both MNL and MXL models.
For mixed logit, we have assumed that all of the preference parameters were random, following normal distributions. In what
follows, we assumed the following form of the utility function of respondent i ∈ I from choosing alternative j ∈ J in choice situation
t ∈ T (time subscript is suppressed):

Uij = βTEL TELij + βTV TVij + βINT INTij + βTELTV TELTVij + βTELINT TELINTij + βTVINT TVINTij + βTELTVINT TELTVINTij + βPRICE PRICEij
i i i i i i i i

+ βONLYMOB ONLYMOBij + ϵij (6)

i

where β is the vector of parameters associated with their respective variables and εij is a random component of utility associated with
alternative j . The interpretation of variables in the choice model is given in Table 3 below. The estimation results – coefficients for
MNL and means and standard deviations of the normally distributed preference parameters for MXL – are reported in Table 4. In
both models, we set ONLYMOB as a baseline utility level. The parameters describe the relative importance (utility) associated with
the attribute levels. Their absolute values do not have an interpretation, but their sign, relative values, and statistical significance
indicate the most important drivers of choice for customers.
While all coefficients are significant in both models, MXL performs significantly better as can be seen from the values of Log-
likelihood and pseudo R2 statistics.16 The most remarkable outcome of both models is that a standalone (naked) fixed-line telephony
brings disutility compared to the baseline alternative of using solely mobile services. This result is expected because fixed telephony
lacking mobility is functionally inferior to mobile voice service and assures us of the correctness of the implemented choice
experiment. Results of MXL show that fixed telephony is redundant as an add-on to the fixed-line broadband and pay TV.
Apparently, subscribers consider TVINT dual-play package as the best option. This finding could explain why the adoption of triple-
play bundle has decreased, as shown in Eurobarometer survey. Altogether, our results imply that customers who use mobile services
at home, will not subscribe to the naked fixed-line telephony – even if it was offered for free. On the managerial grounds, this result
suggests that given increasing popularity of unlimited voice plans, operators can no longer use upsell strategies based on fixed-line
telephony.
A second observation which follows from Table 4 is that both fixed-line broadband and pay TV generate positive utility as
standalone services. We note that the coefficient for INT is largely positive, compared to the baseline mobile services indicating that
broadband is the biggest value generator. This result suggests that mobile Internet available within currently subscribed mobile
services is still not a strong substitute of fixed-line broadband. Perhaps with larger data allowances and increased high-speed mobile
data coverage, the position of fixed-line broadband could be undermined. Finally we note that the utility from TVINT package

16
The log likelihood ratio test rejects the restricted MNL model.

234
M. Sobolewski, T. Kopczewski Telecommunications Policy 41 (2017) 227–241

Table 4
The results of the MNL and MXL models of respondents’ mobile operator choices.

Variables Parameters

MNL MXL

Coefficient Mean Standard deviation

(s.e) (s.e.) (s.e.)

TEL – fixed telephony standalone -1.3856*** -0.1608* 2.1575 ***

(0.1836) (0.5661) (0.5075)
TV – pay TV standalone 0.3705*** 1.7244*** 2.3201***
(0.1401) (0.3751) (0.2351)
INT – fixed Internet standalone 2.7295*** 6.5462*** 3.1473***
(0.1865) (0.4529) (0.3252)
TELTV – telephony+TV dual-play 0.3806*** 0.9254* 3.4928***
(0.1284) (0.4949) (0.4216)
TELINT – telephony+internet dual-play 2.5717*** 6.1233*** 2.7999***
(0.2309) (0.5290) (0.3190)
TVINT – tv+internet dual-play 4.2799*** 10.234*** 2.5427***
(0.2614) (0.5707) (0.2578)
TELTVINT – telephony+tv+internet triple-play 4.3300*** 9.8766*** 2.8111***
(0.2927) (0.6358) (0.4282)
PRICE – monthly subscription fee -0.047*** - 0.1145*** 0.0539***
(0.0031) (0.0069) (0.0036)
Model characteristics
Log-likelihood −3812.709 −2642.30
McFadden's pseudo R2 0.1433 0.4160
AIC/n 2.341 1.629
n (observations) 3264 (13,056) 3264 (13,056)
k (parameters) 8 16

*** ** *
, , Significance at 1%, 5%, 10% level.

Table 5
Definition of dummy variables used as covariates of means of random parameters in MXL model.

Variable name and definition Sample frequency (N=272)

A. User profile
MBBH – has mobile broadband (voice & data plan or dedicated sim) with higher monthly data allowance (more than 1 GB) 68.4%
MBBL – has mobile broadband (voice & data plan or dedicated sim) with lower monthly data allowance (up to 1 GB) 23.9%
UNLMOBV – has unlimited mobile calls 38.6%
FBB – uses fixed-line broadband 84.9%
FTEL – uses fixed-line telephony 39.0%
IPTV – uses paid IP television (cable or xDSL(a)) 29.4%
SATTV – uses paid direct broadcast satellite TV 26.8%
INCOME_LOW – has low monthly net income (less than 375 euros) 25.4%

A. Usage profile
BBVIDEO_HI – high usage of BB for free video (at least few times a week or more) 28.7%
TV_LOWUSE – low usage of TV (up to 3 h weekly) 63.6%
BB_LOWUSE – low usage of BB (up to 2 h daily) 22.1%
TVBB_LOWUSE – low usage of both TV and BB 14.0%

(a)
xDSL denotes a family of Digital Subscriber Line access technologies.

exceeds the sum of utilities from both component services. This is an indication of complementarity between fixed-line broadband
and pay TV, which we discuss in more detail in Section 5.2.

5.1. Exploring sources of preference heterogeneity

Large estimates of the standard deviations (relative to the means) associated with choice characteristics indicate the presence of
substantial unobserved preference heterogeneity in our sample. In MXL framework sources of individual heterogeneity can be
explored by adding a set of choice-invariant covariates as explanatory variables for the means or standard deviations of selected
parameters. Covariate effects might disclose impacts of usage or subscriber profiles on utility from particular bundles. Previous
literature has shown that selected covariates have a significant impact on demand for particular fixed-line and mobile services. For
instance Grzybowski and Verboven (2016) have shown that city or richer households have greater standalone utility for both fixed-
line and mobile services. Likewise, households subscribing fixed-line broadband perceive mobile access as complementary, while

235
M. Sobolewski, T. Kopczewski Telecommunications Policy 41 (2017) 227–241

remaining households treat both services as substitutes.

In what follows, we test for the impact of several dummy variables Z k characterizing user and service profiles of our sample
respondents (see Table 5) on the estimated means of random parameters.
We expect the following hypotheses to hold:

H1. Mobile plan with unlimited calls reduces utility of fixed-line telephony (FMS in voice);.

H2. Higher mobile data allowance decreases utility from fixed-line broadband;.

H3. Low usage intensity of TV or INT decreases their respective utilities;.

H4. Mobile data plan decreases utility from fixed-line broadband (FMS in data);.

H5. High usage intensity of free online streaming video increases utility from fixed-line broadband and decreases utility from pay
TV.
Hypotheses H1-H3 are self-explanatory. Hypothesis H4 describes the relation between mobile and fixed-line broadband
confirmed in recent studies. Grzybowski and Liang (2015) claim that both types of access are complimentary as users browse content
on mobile handsets but complete their online activity at home using fixed broadband. This pattern will likely exist as long as mobile
access is unsuitable for downloading or streaming large video due to lower speeds and data transfer limit providing a justification for
H2. The last statement (H5) is supportive to H4. We expect that massive consumption of free online video should have a positive
impact on the utility of fixed-line broadband, but negative impact on pay TV.
Variables from Table 5 are now assumed to influence means of selected parameters in utility function given in Eq. (6). For
example, parameter βTEL_i has now the following distribution:
⎛ ⎞
βTEL ~ N ⎜βTEL + βMBBH MBBHi +…β z k Zik…+βTVBB _LOWUSE TVBB _LOWUSEi, 2
σTEL ⎟
i ⎝ TEL TEL TEL ⎠

Similar expressions can be specified for model parameters representing utilities of other fixed-line packages. To maintain clarity,
we do not analyze heterogeneity in the means for the two least popular dual-play packages, namely TELINT and TELTV. The results
of this model are reported in Table 6. The interpretation of the coefficients given in panel A should now be made together with the
coefficients in panel B – covariates of means of normally distributed random parameters. Based on model characteristics provided in
panel C, we note that introduction of mean covariate effects further increased model performance as indicated by changes to pseudo-
R2, and Akaike information criterion.
The model shows clear patterns of preference heterogeneity. The mean utility from naked fixed-line telephony (TV) is sharply
lower among users of unlimited mobile voice plans as indicated by negative value of coefficient for covariate UNLMOBV. Standalone
pay TV offer is a more preferred option among users with lower intensity of broadband use (BB_LOWUSE) and among users of
mobile internet, regardless of the size of monthly data allowance (coefficients for covariates MBBH and MBBL are both positive and
significant). On the other hand utility from naked pay TV is lower among individuals with low usage intensity of TV (TV_LOWUSE).
Low income respondents prefer to economize and adopt TV in dual-play bundle, but avoid triple-play package, as suggested by
negative coefficient of INCLOME_LOW for TELTVINT and TV. Interestingly, high consumption of free online video does not reduce
utility from pay TV. This result suggests that free content from online operators such as YouTube is not substitutable for traditional
formats offered by cable or xDSL operators. However, the spread of Netflix or video on demand (VOD) platforms which offer paid
online content, will likely affect preferences for traditional Pay TV format.17
Preferences for naked fixed-line broadband are dependent as well on the possession and usage intensity of other services. First,
we note that utility of INT is lower among users of fixed-line telephony. This group is on average older, less educated and less
Internet-oriented. Stronger preference for naked fixed-line broadband occurs among individuals with low intensity of TV use and
high intensity of broadband use for video streaming. We also note the positive effect of IPTV dummy on mean utility from naked
broadband. This is presumably artefactual effect caused by marketing practice of incumbent xDSL operator who added basic IPTV
package for free. Interestingly our data does not show any statistically significant effect of mobile internet on naked fixed-line
broadband. However we observe the negative effect of MBBH covariate on utility from packages containing fixed-line broadband
(TVINT and TELTVINT). This result suggests certain degree of substitutability between fixed-line access and mobile plans with
larger data allowance. Partially substitutability is supported by another finding showing positive effect of MBBH on utility from
naked TV. This suggest that TVINT dual-play package competes with combination of naked pay TV and mobile data plan. Moreover,
mobile plan with lower data allowance (MBBL) does not reduce utility from fixed broadband (both naked and bundled). This result
implies that mobile data plan with low transfer limits is not substitute to fixed-line access, but rather an independent service
dedicated to different online activities, as suggested in previous studies.
With regards to the preferences for bundles, our model shows that both TVINT and TELTVINT packages are less preferred by
subscribers of mobile data plans with larger allowance, as well as respondents with low usage intensity of TV and broadband
(TV_LOWUSE, TV & BB_LOWUSE). On the other hand utility of each package is greater among users of pay TV as indicated by
positive signs for IPTV and SATTV dummies. In the case of TVINT dual-play adoption of fixed-line broadband or unlimited mobile
voice plan has positive effect on utility.

17
Unfortunately, we are unable to test this hypothesis with our data.

236
M. Sobolewski, T. Kopczewski Telecommunications Policy 41 (2017) 227–241

Table 6
The results of MXL model with covariates of means of normally distributed random parameters.

A. Parameters of MXL
Mean (s.e.) Standard deviation (s.e.)
TEL – fixed telephony standalone 0.613 1.811***
(0.906) (0.667)
TV – pay TV standalone -0.458 3.297***
(1.044) (0.381)
INT – fixed Internet standalone 5.659*** 2.590***
(0.780) (0.243)
TELTV – telephony+TV dual-play 0.686* 3.141***
(0.391) (0.279)
TELINT – telephony+internet dual-play 7.092*** 1.799***
(0.465) (0.229)
TVINT – tv+internet dual-play 7.363*** 2.502***
(0.962) (0.227)
TELTVINT – tel+tv+internet triple-play 9.501*** 2.080***
(0.970) (0.234)
PRICE – monthly subscription fee -0.112*** 0.051***
(0.007) (0.003)
B. Coefficients with (s.e.) of covariates of means of random parameters
Covariates TEL TV INT TVINT TELTVINT
MBBH – resp. uses mobile internet with higher data allowance -0.096 2.800*** -0.283 -0.876* -1.007**
(0.608) (0.716) (0.451) (0.524) (0.475)
MBBL – resp. uses mobile internet with lower data allowance -1.593* 1.789** 0.625 0.464 0.275
(0.863) (0.828) (0.518) (0.586) (0.590)
UNLMOBV – resp. has unlimited mobile voice -1.452** 0.253 -0.060 0.855** 1.401***
(0.706) (0.489) (0.394) (0.427) (0.503)
FBB – resp. uses fixed-line BB -0.136 0.930 -0.313 2.552*** -0.063
(0.740) (0.662) (0.525) (0.573) (0.512)
FTEL – resp. uses fixed-line telephony 0.686 -0.984* −1.220*** -1.080 0.805*
(0.548) (0.502) (0.412) (0.420) (0.430)
IPTV – resp. uses paid IPTV (cable or xDSL) -0.907 1.787** 1.663*** 3.151*** 2.693***
(0.813) (0.601) (0.489) (0.502) (0.458)
SATTV – resp. uses paid Satellite TV 0.323 1.100* 0.104 2.445*** 2.229***
(0.619) (0.561) (0.463) (0.532) (0.474)
INCOME_LOW – resp. has low net income -1.095 -1.425** 0.850* -0.819 -0.978**
(0.852) (0.602) (0.489) (0.529) (0.453)
TV_LOWUSE – resp. low usage of TV -1.194 −2.557*** 1.250*** −0.801* −1.143***
(0.760) (0.540) (0.467) (0.439) (0.434)
BB_LOWUSE – resp. has low usage of BB 1.051 1.718** 0.035 1.535** 0.792
(1.031) (0.751) (0.658) (0.663) (0.643)
TV & BB_LOWUSE – resp. has low usage of TV and BB 0.476 -0.067 -0.902* −2.203*** −1.425*
(1.190) (1.101) (0.866) (0.905) (0.863)
BBVIDEO_HI – resp. has high usage of BB for video -0.568 -0.661 1.160*** 1.160*** 0.134
(0.718) (0.523) (0.423) (0.438) (0.410)
C. Model characteristics
Log-likelihood −2528.07
McFadden's pseudo R2 0.441
AIC/n 1.596
n (observations) 3264 (13,056)
k (parameters) 76

*** ** *
, , Significance at 1%, 5%, 10% level.

Taken as a whole our analysis offers several interesting implications. First, we obtain evidence of partial substitutability between
fixed-line and mobile broadband access. It is limited to mobile data plans with larger transfer limits. This fixed-to-mobile
substitution in data services is likely to gain on strength with gradual increase of data capacity in mobile networks. Second,
introduction of unlimited mobile voice plans sharply decreases utility from fixed-line telephony, below the baseline level set for
mobile telephony. This implies that a widespread adoption of unlimited tariffs will eventually crowd out fixed-line telephony. Third,
the demand for fixed-line broadband both naked and bundled with TV is driven by the use of online video services, which has
positive implications for broadband providers. On the other hand the prospects for broadcasters related with an impact on online
streaming are less promising. The effect of online video streaming on naked pay TV is (as yet) insignificantly negative. The latter
finding might foretell an increasing competition of OTT services such as Hulu or Netflix, which offer a similar formats and quality as
traditional broadcasters. Fourth, intensities of TV and broadband use are also informative for explaining preference heterogeneity.
For example, naked fixed-line broadband seems to be an attractive service for subscribers with low use of pay TV and high use
broadband for free on-line content. Interestingly the utility from fixed-line broadband is independent of its usage intensity.
Apparently, fixed-line broadband is considered as an essential service by all users, regardless whether they use it for time consuming

237
M. Sobolewski, T. Kopczewski Telecommunications Policy 41 (2017) 227–241

Table 7
Mean median WTP based on MXL and means of self-reported valuations in PLN.

Bundle WTP based on MXL estimates Self-reported valuations

Mean Median [PLN] 95% c.i. Mean [PLN]

TEL -9.80 -19.68 to -0.43 16.47

TV 14.48 8.60–20.10 32.61
INT 55.45 50.23–60.54 51.5
TELTV 7.74 -0.48–15.72 34.64
TELINT 51.95 45.35–58.45 49.43
TVINT 87.25 83.17–91.51 68.93
TELTVINT 84.13 77.91–90.21 72.99

activities such as streaming and gaming or sporadically for banking and communication. Altogether, our findings fully confirm the
first two hypotheses and partially confirm H3–H5.

5.2. Bundle valuations

In Table 7 we present monthly median WTP estimates for fixed-line bundles and their components based on MXL model.18
Together with MXL-WTP we provide means of self-reported valuations of sample respondents.
Our results bring a couple of interesting observations. First, MXL-WTP values for TEL, TELTV, TELINT and TELTVINT indicate
that fixed-line telephony indeed does not generate any positive value for subscribers in our sample neither as a standalone service
nor a component of dual-play and triple-play bundles. Since the no-choice alternative in our model implied using solely currently
subscribed mobile services, this finding confirms fixed-to-mobile substitution of voice telephony in money-metric terms.
Secondly, the sum of WTP for INT and TV is smaller (69.94 PLN) than WTP for dual-play bundle (87.25 PLN). Thus, MXL-WTP
estimates illustrate sharp complementarity of fixed-line broadband and pay TV. At first glance, this result looks striking. Large
availability of free online video would suggest the opposite effect (i.e. substitutability of TV and INT). This is however not a case as
indicated by our findings from Section 5.1. One possible reason of complementarity appeals to technical integration, i.e. access to
video-on-demand libraries or ability to watch subscribed TV channels on personal mobile devices. Another explanation relates to
single provider effect and related convenience benefit. A third possibility is that free-of-charge OTT video services are simply not
substitutes for pay TV due to much lower quality and no overlap of legal content due to strict copyright infringement policy.19
Complementarity of broadband and pay TV is also found by Liu, Chintagunta, and Zhu (2010) within a discrete choice model on a
panel data of US households. Our paper documents similar findings for polish subscribers.
Third, Table 7 indicates substantial differences between valuations obtained from direct and indirect elicitation formats. With the
exception of two packages (INT and TVINT), self-reported valuations fall off the confidence intervals of MXL-WTP estimates.
Consumers sharply overestimate values of TV, TEL and TELTV packages in direct declarations. On the other hand self-reported
valuations for INT and TVINT are smaller in comparison to choice-based reservation prices, but the difference is not statistically
significant. Another discrepancy between direct and indirect valuations is reflected by subadditivity of self-reported valuations for TV
and INT. This outcome suggests that self-reported valuations are likely biased by bundle discounts applied by service providers.
Hence, direct valuation may indicate how much a consumer would like to pay or what is his perception of a fair price rather than
actual reservation price. Such sensitivity to various heuristics undermines credibility of direct valuations.
Fig. 2 illustrates cumulative distributions of MXL-WTP for particular fixed-line bundles, simulated with Krinsky and Robb
method together with implied demand functions and empirical distributions of valuations self-reported in the survey.20 Direct
valuations of TEL, TV and TELTV seem to be overstated in the whole range of values, compared to choice-based WTP. On the other
hand respondents seem to undervalue naked broadband and TVINT in declarations. This outcome suggests that the revenues from
pay TV might be preserved by integrating it with broadband.

6. Summary and conclusions

This paper derives willingness-to-pay distributions for the three most popular fixed-line telecommunication services, namely
telephony, broadband internet and pay TV in a stated preference discrete choice experiment. Our framework encompasses specificity
of telecommunications services, namely: possible substitutability and complementarities between bundle components as well as the
impact of mobile and over-the-top video services. To our knowledge, this is the first attempt to elicit choice-based valuations for
three-component bundles in telecommunications. Contrary to choice-based WTP estimates obtained from mixed logit model, self-

18
WTP estimates have been obtained based on the estimates from Table 4, using Krinsky and Robb simulation with 105 random draws at each stage (see Section
4.2 for details).
19
For example YouTube and major file sharing providers have policies for removing illicit content. We thank one of anonymous Reviewers for paying our attention
to this issue.
20
For a given cumulative WTP distribution F (x ) , a demand function is calculated as 1 − F (x ).

238
M. Sobolewski, T. Kopczewski Telecommunications Policy 41 (2017) 227–241

Fig. 2. Simulated cumulative distributions of choice-based WTP and demands.

reported valuations are exposed to various heuristics, which lead to biased demand assessment. Researchers and practitioners
should, therefore, be very careful when using values self-reported by respondents.
This paper explores sources of preference heterogeneity with usage and service profiles of our respondents. We have found that
the use of mobile broadband does not reduce utility from fixed-line broadband if data allowance is small. However for higher transfer
limits, both types of access are perceived as imperfect substitutes. The use of free online video services showed negative effect on
utility from pay TV. This result is however insignificant and must be taken with caution. Despite this, broadcasters and media
platform operators are not in a safe position as they will increasingly face a competitive threat from over-the-top operators such as
Netflix or Hulu, who offer paid content of similar format to traditional pay TV providers.
Broadband is the biggest value generating service in fixed networks. It also offers a potential for additional value creation from
integration with pay TV, as indicated for example by super-additivity of valuation for TVINT dual-play. We note that out of seven
combinations of fixed line packages, three types seem to be the most promising: that is naked broadband, TVINT dual-play and
naked TV. The latter one is frequently chosen by subscribers who rely solely on mobile technology with respect to voice and data
services. Fixed-line voice is already perceived as a redundant service, because of widespread use of unlimited mobile voice plans.
Although this outcome is derived from non-representative sample it confirms earlier findings of fixed-to-mobile substitution
literature in the group of younger respondents, often classified as digital natives.
Altogether the study proves that preference elicitation studies are helpful in disclosing upcoming changes in consumer behavior
and can serve as early indicators of shifts in market trends related to substitution between fixed and mobile services.

239
M. Sobolewski, T. Kopczewski Telecommunications Policy 41 (2017) 227–241

Acknowledgements

We are grateful to Mikołaj Czajkowski for helpful advice on experimental design for this study. Financial support from the Polish
National Science Center (grant no. 2013/09/B/HS4/02728) is gratefully acknowledged. We thank two anonymous referees for
detailed and relevant comments. All errors remain ours.

Appendix

We set priors for respective utility parameters based on declared willingness to pay for all combinations of packages obtained
from 6 subjects during the pretest phase. The priors have been calculated by multiplying means of collected reservation prices by
price coefficient (set to minus 0.01). We have utilized those priors to obtain (Bayesian) efficient design using Ngene package. Because
our chosen optimization criterion (C-error) is implemented only for multinomial logit we do not randomize priors. However to
account for uncertainty about common price sensitivity, we assume that it is uniformly distributed between [−0.015, −0,005]. The
optimized design is presented below in Table A1. Respondents were randomly assigned to one of three blocks.

Table A1
Final design for the study.

Choice situation Alt1a Price1b Alt2a Price2b Alt3a Price3b Alt4c block

1 1 10 3 36 7 74 nc 1
2 2 31 3 56 7 95 nc 1
3 3 82 2 61 5 59 nc 1
4 6 74 7 95 5 31 nc 1
5 4 95 5 41 2 51 nc 1
6 5 23 3 49 1 23 nc 1
7 2 28 1 23 6 64 nc 1
8 5 36 6 82 7 100 nc 1
9 5 54 2 51 3 79 nc 1
10 3 51 6 54 4 72 nc 1
11 7 69 6 43 1 10 nc 1
12 4 90 5 33 2 41 nc 1
13 7 79 4 74 3 43 nc 2
14 7 87 5 20 1 18 nc 2
15 7 100 6 79 2 38 nc 2
16 1 25 3 46 5 20 nc 2
17 1 15 7 77 3 36 nc 2
18 3 49 4 72 1 15 nc 2
19 6 64 7 90 3 54 nc 2
20 2 38 4 87 7 97 nc 2
21 3 46 1 18 6 61 nc 2
22 5 41 4 92 6 82 nc 2
23 4 59 1 13 7 69 nc 2
24 6 92 2 59 5 49 nc 2
25 4 67 1 15 2 25 nc 3
26 1 13 4 67 6 56 nc 3
27 4 85 5 28 6 77 nc 3
28 2 43 7 100 5 33 nc 3
29 3 61 2 38 4 87 nc 3
30 5 33 7 97 3 67 nc 3
31 6 77 3 64 4 90 nc 3
32 7 56 1 10 4 46 nc 3
33 2 18 7 85 1 13 nc 3
34 1 20 6 69 2 28 nc 3
35 7 97 5 25 4 85 nc 3
36 6 72 2 31 7 92 nc 3

a
Dummy coding of packages: 1=TEL, 2=TV, 3=INT, 4=TV+INT, 5=TV+TEL, 6=INT+TEL, 7=TV+INT+TEL.
b
Price is treated as continuous variable from range [10,100].
c
‘No choice’ alternative.

240
M. Sobolewski, T. Kopczewski Telecommunications Policy 41 (2017) 227–241

References

Adams, W. J., & Yellen, J. L. (1976). Commodity bundling and the burden of monopoly. The Quarterly Journal of Economics, 475–498.
Backhaus, K., Wilken, R., Voeth, M., & Sichtmann, C. (2005). An empirical Comparison of methods to measure willingness to pay by examining the hypothetical bias.
International Journal of Market Research, 47(5), 543–562.
Barth, A.-K., & Heimeshoff, U. (2012). How large is the magnitude of fixed-mobile call substitution? Empirical evidence from 16 European countries.
Telecommunications Policy, 38(8–9), 771–782.
Ben-Akiva, M., & Gershenfeld, S. (1998). Multi-featured products and services: Analysing pricing and bundling strategies. Journal of Forecasting, 17(3–4), 175–196.
Ben-Akiva, M., & Lerman, S. R. (1985). Discrete choice analysis: Theory and application to travel demand Cambridge, MA: MIT Press.
Bliemer, M. C., & Rose, J. M. (2013). Confidence intervals of willingness-to-pay for random coefficient logit models. Transportation Research Part B:
Methodological, 58, 199–214.
Carare, O., McGovern, C., Noriega, R., & Schwarz, J. (2015). The willingness to pay for broadband of non-adopters in the US: Estimates from a multi-state survey.
Information Economics and Policy, 30, 19–35.
Carson, R. T., Flores, N. E., Martin, K. M., & Wright, J. L. (1996). Contingent valuation and revealed preference methodologies: Comparing the estimates for quasi-
public goods. Land Economics, 72(1), 80–99.
Chen, Y., & Riordan, M. H. (2013). Profitability of product bundling. International Economic Review, 54(1), 35–57.
Czajkowski, M., & Sobolewski, M. (2016). How much do switching costs and Personal network effects contribute to Consumer lock-in in mobile telephony?
Telecommunications Policy, 40(9), 855–869.
Digital TV Research (2013). Triple-Play Forecasts: toc. Retrieved from : 〈https://www.digitaltvresearch.com/ugc/Triple-play%20Forecasts%20TOC
%202013%20pdf_toc_87.pdf〉.
Eckalbar, J. C. (2010). Closed‐form solutions to Bundling problems. Journal of Economics & Management Strategy, 19(2), 513–544.
European Commission (2014). Special Eurobarometer 414. E-Communications and Telecom Single Market Household Survey. Retrieved from: 〈http://ec.europa.
eu/information_society/newsroom/cf/dae/document.cfm?Doc_id=2629〉.
European Commission (2015). Special Eurobarometer 438. E-Communications and the Digital Single Market. Retrieved from: 〈http://ec.europa.eu/
COMMFrontOffice/publicopinion/index.cfm/Survey/getSurveyDetail/instruments/SPECIAL/surveyKy/2062〉.
Greene, W. H. (2011). Econometric analysis (7 ed.) Upper Saddle River, NJ: Prentice Hall.
Greene, W. H., & Hensher, D. A. (2007). Heteroscedastic control for random coefficients and error components in mixed logit. Research Part E Logistics and
Transportation Review, 43(5), 610–623.
Grzybowski, L. (2014). Fixed-to-mobile substitution in the European Union. Telecommunications Policy, 38(7), 601–612.
Grzybowski, L., & Liang, J. (2015). Estimating demand for fixed-mobile bundles and switching costs between tariffs. Information Economics and Policy, 33, 1–10.
Grzybowski, L., Nitsche, R., Verboven, F., & Wiethaus, L. (2014). Market definition for broadband internet in Slovakia–Are fixed and mobile technologies in the same
market? Information Economics and Policy, 28, 39–56.
Grzybowski, L., & Verboven, F. (2016). Substitution between fixed-line and mobile access: The role of complementarities. Journal of Regulatory Economics, 49(2),
113–151.
Hensher, D., & Greene, W. (2003). The mixed logit model: The state of practice. Transportation, 30(2), 133–176.
Hole, A. R. (2007). Modelling heterogeneity in patients' preferences for the attributes of a general Practitioner appointment. Journal of Health Economics, 27(4),
1078–1094.
Jedidi, K., Jagpal, S., & Manchanda, P. (2003). Measuring heterogeneous reservation prices for product bundles. Marketing Science, 22(1), 107–130.
Klein, A., & Jakopin, N. (2014). Consumers' willingness-to-pay for mobile telecommunication service bundles. Telematics and Informatics, 31(3), 410–421.
Krämer, J. (2009). Bundling vertically differentiated communications services to leverage market power. Info-The Journal of Policy, Regulation and Strategy for
telecommunications, 11(3), 64–74.
Krinsky, I., & Robb, A. L. (1986). On approximating the statistical properties of elasticities. The Review of Economics and Statistics, 715–719.
Liu, H., Chintagunta, P. K., & Zhu, T. (2010). Complementarities and the demand for home broadband internet services. Marketing Science, 29(4), 701–720.
Louviere, J. J., Hensher, D. A., & Swait, J. D. (2006). Stated choice methods: Analysis and applications Cambridge: Cambridge University Press.
McAfee, R. P., McMillan, J., & Whinston, M. D. (1989). Multiproduct monopoly, commodity bundling, and correlation of values. The Quarterly Journal of
Economics, 371–383.
McFadden, D. (1974). Conditional logit analysis of qualititative choice behaviour. , in: Zarembka, P. (Ed.). (1974). Frontiers in econometrics . New York, NY:
Academic Press, 105–142.
Miller, K. M., Hofstetter, R., Krohmer, H., & Zhang, Z. J. (2011). How should consumers' willingness to pay be measured? An empirical comparison of state-of-the-art
approaches. Journal of Marketing Research, 48(1), 172–184.
Prince, J., & Greenstein, S. (2014). Does service bundling reduce churn? Journal of Economics & Management Strategy, 23(4), 839–875.
Sándor, Z., & Wedel, M. (2001). Designing conjoint choice experiments using managers’ prior beliefs. Journal of Marketing Research, 38(4), 430–444.
Scarpa, R., & Rose, J. M. (2008). Design efficiency for non-market valuation with choice modelling: How to measure it, what to report and why. Australian Journal of
Agricultural and Resource Economics, 52(3), 253–282.
Schmalensee, R. (1984). Gaussian demand and commodity bundling. Journal of Business, S211–S230.
Srinuan, P., Srinuan, C., & Erik, B. (2014). An empirical analysis of multiple services and choices of consumer in the Swedish telecommunications market.
Telecommunications Policy, 38(5–6), 449–459.
Stigler, G. J. (1963). United States v. Loew's Inc.: A note on block-booking. Sup Ct Rev, 152.
Street, D. J., & Burgess, L. (2007). The construction of optimal stated choice experimentsTheory and Methods, 647 Hoboken, New Jersey: John Wiley Sons.
Stremersch, S., & Tellis, G. J. (2002). Strategic bundling of products and prices: A new synthesis for marketing. Journal of Marketing, 66(1), 55–72.
Train, K. E. (2009). Discrete choice methods with simulation New York: Cambridge University Press.
Üner, M. M., Güven, F., & Cavusgil, S. T. (2015). Bundling of telecom offerings: An empirical investigation in the Turkish market. Telecommunications Policy, 39(1),
53–64.
Venkatesh, R., & Mahajan, V. (2009). The design and pricing of bundles: A review of normative guidelines and practical approaches. , in: Vithala, R. (Ed.). (2009).
Handbook of pricing research in marketing . . Edward Elgar Publishing, 232–257.
Vossler, C. A., & Watson, S. B. (2013). Understanding the consequences of consequentiality: Testing the validity of stated preferences in the field. Journal of
Economic Behavior and Organization, 86, 137–147.
Wang, T., Venkatesh, R., & Chatterjee, R. (2007). Reservation price as a range: An incentive-compatible measurement approach. Journal of Marketing Research,
44(2), 200–213.
Wertenbroch, K., & Skiera, B. (2002). Measuring consumers’ willingness to pay at the point of purchase. Journal of Marketing Research, 39(2), 228–241.
Wu, S.-Y., Hitt, L. M., Chen, P.-Y., & Anandalingam, G. (2008). Customized bundle pricing for information goods: A nonlinear mixed-integer programming approach.
Management Science, 54(3), 608–622.
Yang, M. (2013). Churn management and policy: Measuring the effectiveness of fixed-mobile bundling on mobile subscriber retention. Journal of Media Economics,
26(4), 170–185.
Yang, B., & Ng, C. (2010). Pricing problem in wireless telecommunication product and service bundling. European Journal of Operational Research, 207(1),
473–480.

241