Consumer preferences for mobile phone service in the U.S.: an application of efficient design on conjoint analysis

Abstract

The wide commercial success of certain mobile phones, such as Apple‘s iPhone and RIM‘s Blackberry, was the motivation behind this study to examine empirically what drives the demand for mobile service bundles. If casual observation is an accurate indicator, consumers make their mobile purchasing decisions based solely on the type of mobile phone that mobile service providers are offering at the time as part of a bundle of services. This, in turn, raises the question of whether service bundle components, other than the mobile phone, matter to consumers. In light of increased competition and saturation in the U.S. mobile sector, gaining a deeper understanding of consumer choice is critical not only for the development of effective market strategies but also for policymaking. As governmental agencies take a closer look at competition and the need or lack thereof of regulation in the mobile sector, it is crucial to understand how consumers purchase mobile service as this may very well form the basis of new regulations and public policies. Surprisingly, although there is a large literature addressing various aspects of mobile demand, no prior study has examined this topic from a mobile service bundle perspective.The present study uses data from an online stated-preference survey with a conjoint analysis component. The design for the conjoint analyses incorporates efficient survey design, which promises most accurate parameter estimates. It is the first application of efficient survey design theory to telecommunication services. It is also one of the first practical applications of this innovative concept. In these trade-off exercises, 503 survey respondents ranked three mobile service plan alternatives, each described via 10 service attributes. Survey respondents completed six such exercises. A thorough quality review of the survey results revealed 14 invalid survey responses and survey respondent fatigue in the last two choice situations. After eliminating the 14 invalid responses, the resulting data were fit to several versions of the multinomial exploded logit model. Using likelihood ratio indices and hypotheses tests, such as the likelihood-ratio test, the Wald test, and the Hausman test, to determine the best model for this study, the final model selected was a multinomial mixed exploded logit model with 10 lognormal distributed and two fixed parameters. This model provides direct insight into the demand determinants for mobile service bundles. It reveals demand elasticities and calculates the consumers‘ maximum willingness to pay for specific bundle components.The fitted model reveals several interesting econometric, competitive, and public policy findings. First, applying D-efficient survey design requires a priori information on the final model‘s specification and the signs and sizes of its coefficients. Data from a pilot survey fitted to a multinomial logit model generate the necessary a priori proxies. The design matrix is D-optimized relative to this a priori model. Any deviation from the pilot model‘s specifications and its coefficient priors jeopardizes the optimality of the design matrix. A test was derived to measure whether the optimized design matrix retained its optimality when evaluated under the final model instead of the pilot model. In the present study, the final model specifications and coefficients deviate sufficiently from the a priori proxy to render the optimized design matrix no more or less efficient than a design matrix randomly created. Hence, no benefits from D-optimization carried through to the final model. With perfect foresight, however, D-optimality could have decreased the design matrix‘s D-error by 83%, thereby significantly increasing the model‘s accuracy. This practical application of D-efficient survey design illustrates that further research in efficient design needs to address how the benefits from D-optimization can be retained.In terms of competition, the fitted model explores several competitive strategies, simulating market share gains and losses from changes in attribute levels and calculating demand elasticities for specific bundle components. This analysis reveals that only certain pricing strategies are effective. It also demonstrates that a combinatorial strategy might be most effective. Specifically, decreasing mobile phone prices, increasing term lengths, and increasing the monthly recurring charge increases subscriber revenue in addition to gaining market share.In terms of public policy, the study finds that regulators must examine market behavior and alleged market failures in terms of service bundles. Considering individual bundle attributes on a standalone basis, which is currently the common practice, yields incorrect results. Finally, the fitted model highlights the importance of making additional radio spectrum available to mobile service providers