Abstract
Social network analysis has been a topic of regular interest in the marketing discipline. Previous studies have largely focused on similarities in product/brand choice decisions within the same social network, often in the context of product innovation adoption. Not much is known, however, about the importance of social network effects once customers have been acquired. Using the customer base of a telecommunications company, our study analyzes network autocorrelation in the distribution of customer-level revenue within a social network. Our results indicate a significant and substantial degree of positive network autocorrelation in customer-level revenue. High (low) revenue customers therefore tend to be primarily related to other high (low) revenue clients. Furthermore, we show that approximating communicative proximity by spatial proximity leads to a substantial underestimation of these effects.
Similar content being viewed by others
Notes
Autocorrelation refers to the correlation of a variable (e.g., customer-level revenue) with itself. It is a term frequently used in the context of time-series analysis where it describes the correlation between two values of the same variable measured at different points in time (i.e., temporal autocorrelation). It is, however, also used in spatial statistics to represent the correlation between two values of the same variable measured at different locations (i.e., spatial autocorrelation) and social network analysis.
We calculated the distance between all pairs of actors and subsequently normalized those distances by dividing them by the maximum sample distance. Inverse geographical proximity was then defined as 1 − normalized distance.
Specifically, we used the following functions: moran.test and geary.test to calculate Moran’s I and Geary’s C; lm.morantest to calculate Moran’s I for regression residuals and errorsarlm to estimate the spatial simultaneous autoregressive error model in the context of the first robustness check; and joincount.test to calculate join-count statistics in the context of the second robustness check.
While this was possible for virtually all actors within sample B (only 55 out of 19,668 actors showed missing values with respect to the age variable), we were only able to obtain such information for roughly 42% of the 6,681 actors within sample A. The results for the latter case therefore need to be interpreted with caution.
See http://www.appliedgeographic.com/mosaic.html for additional details on the Mosaic typology.
Details on this analysis can be obtained from the author on request.
The three-part tariff structure of the mobile phone operator induces a relationship between customer-level revenue and service plan choice. Mean customer-level revenue for all clients within the same service plan ranges from 0.23 to 1.72 for the 15 service plans analyzed. Nevertheless, due to substantial mobile phone usage outside of the monthly service plan allowance, the standard deviations of the mean are significant (between 0.23 and 0.98). This leads to the fact that although some service plans are associated with strictly larger or smaller customer-level revenue than others, many service plans overlap in terms of customer-level revenue. Service plan choice and customer-level revenue are therefore two distinct measures of post-acquisition customer behavior.
The mobile phone company we collaborated with is based in Europe, where the predominant billing scheme is Calling Party Pays. This implies that the mobile subscriber does not pay for incoming calls but instead the calling party pays for those calls. In order to notify the calling customer that s/he has called a number for which there will be a different tariff, mobile numbers in Europe are usually dedicated to specific blocks. This made it straightforward to generate a set of random numbers which had a reasonably high chance of corresponding with actual mobile phone numbers.
It was necessary to define a cut-off in terms of call duration to eliminate numbers from our analysis that have been called only rarely and that, hence, are unlikely to represent true friends. In line with the well-established 80/20 law, we assumed that friends should account for 80% of total call duration. We subsequently tested a range of potential thresholds between 1% and 5% to identify to what extent they fulfilled this criterion. While a 1% cut-off resulted in friends accounting for 84% of total call duration, a 2% cut-off would have resulted in 73% and a 3% cut-off in 64%. Based on these results, we decided to apply a 1% threshold.
References
Ahuja, M. K., Galletta, D. F., & Carley, K. M. (2003). Individual centrality and performance in virtual R&D groups: An empirical study. Management Science, 49(1), 21–38.
Baumgartner, H., & Pieters, R. (2003). The structural influence of marketing journals: A citation analysis of the discipline and its subareas over time. Journal of Marketing, 67(2), 123–139.
Bronnenberg, B. J., & Mahajan, V. (2001). Unobserved retailer behavior in multimarket data: Joint spatial dependence in market shares and promotion variables. Marketing Science, 20(3), 284–299.
Campbell, D., & Frei, F. (2004). The persistence of customer profitability: Empirical evidence and implications from a financial services firm. Journal of Service Research, 7(2), 107–123.
Crosier, K., & Erdogan, B. Z. (2001). Advertising complainants: Who and where are they? Journal of Marketing Communications, 7(2), 109–120.
Fader, P. S., Hardie, B. G. S., & Lee, K. L. (2005a). Counting your customers the easy way: An alternative to the Pareto/NBD model. Marketing Science, 24(2), 275–284.
Fader, P. S., Hardie, B. G. S., & Lee, K. L. (2005b). RFM and CLV: Using iso-value curves for customer base analysis. Journal of Marketing Research, 42(4), 415–430.
Farr, M., & Webber, R. (2001). MOSAIC: From an area classification system to individual classification. Journal of Targeting, Measurement and Analysis for Marketing, 10(1), 55–65.
Feld, S. L., & Carter, W. C. (2002). Detecting measurement bias in respondent reports of personal networks. Social Networks, 24(4), 365–383.
Frenzen, J. K., & Davis, H. L. (1990). Purchasing behavior in embedded markets. Journal of Consumer Research, 17(1), 1–12.
Garber, T., Goldenberg, J., Libai, B., & Muller, E. (2004). From density to destiny: Using spatial dimension of sales data for early prediction of new product success. Marketing Science, 23(3), 419–428.
Geary, R. C. (1954). The continguity ratio and statistical mapping. The Incorporated Statistician, 5(3), 115–145.
Goldenberg, J., Libai, B., & Muller, E. (2001). Talk of the network: A complex systems look at the underlying process of word-of-mouth. Marketing Letters, 12(3), 211–223.
Goldenberg, J., Libai, B., Moldovan, S., & Muller, E. (2007). The NPV of bad news. International Journal of Research in Marketing, 24(3), 186–200.
Goldenberg, J., Han, S., Lehmann, D. R., & Hong, J. W. (2009). The role of hubs in the adoption process. Journal of Marketing, 73(2), 1–13.
Hill, S., Provost, F., & Volinsky, C. (2006). Network-based marketing: Identifying likely adopters via consumer networks. Statistical Science, 21(2), 256–276.
House, J. S., & Mortimer, J. (1990). Social structure and the individual: Emerging themes and new directions. Social Psychology Quarterly, 53(2), 71–80.
Kelman, H. C. (1961). Processes of opinion change. Public opinion Quarterly, 25(1), 57–78.
Krackhardt, D., & Porter, L. W. (1985). When friends leave: A structural analysis of the relationship between turnover and stayers’ attitudes. Administrative science quarterly, 30(2), 242–61.
Krackhardt, D., & Porter, L. W. (1986). The snowball effect: Turnover embedded in communication networks. Journal of Applied Psychology, 71(1), 50–55.
Lambrecht, A., Seim, K., & Skiera, B. (2007). Does uncertainty matter? Consumer behavior under three-part tariffs. Marketing Science, 26(5), 698–710.
Levin, D. Z., & Cross, R. (2004). The strength of weak ties you can trust: The mediating role of trust in effective knowledge transfer. Management Science, 50(11), 1477–1490.
Manchanda, P., Xie, Y., & Youn, N. (2008). The role of targeted communication and contagion in product adoption. Marketing Science, 27(6), 961–976.
Manski, C. F. (1993). Identification of endogenous social effects: The reflection problem. Review of Economic Studies, 60(3), 531–542.
Marin, A. (2004). Are respondents more likely to list alters with certain characteristics? Implications for name generator data. Social Networks, 26(4), 289–307.
Mathwick, C., Wiertz, C., & de Ruyter, K. (2008). Social capital production in a virtual P3 community. Journal of Consumer Research, 34(6), 832–849.
McPherson, M., Smith-Lovin, L., & Cook, J. M. (2001). Birds of a feather: Homophily in social networks. Annual Review of Sociology, 27(1), 415–444.
Mizruchi, M. S. (1994). Social network analysis: Recent achievements and current controversies. Acta Sociologica, 37(4), 329–343.
Money, R. B., Gilly, M. C., & Graham, J. L. (1998). Explorations of national culture and word-of-mouth referral behavior in the purchase of industrial services in the United States and Japan. Journal of Marketing, 62(4), 76–87.
Moran, P. A. P. (1948). The interpretation of statistical maps. Journal of the Royal Statistical Society: Series B (Methodological), 10(2), 243–251.
Moran, P. A. P. (1950). Notes on continuous stochastic phenomena. Biometrika, 37(1/2), 17–23.
Moreno, J. L. (1941). Foundations of sociometry: An introduction. Sociometry, 4(1), 15–35.
Nair, H. S., Manchanda, P, & Bhatia, T. (2010). Asymmetric social interactions in physician prescription behavior: the role of opinion leaders. Journal of Marketing Research (in press).
Ord, K. (1975). Estimation methods for models of spatial interaction. Journal of the American Statistical Association, 70(349), 120–126.
Powell, J., Tapp, A., & Sparks, E. (2007). Social marketing in action—Geodemographics, alcoholic liver disease and heavy episodic drinking in Great Britain. International Journal of Nonprofit and Voluntary Sector Marketing, 12(3), 177–187.
Reagans, R. (2005). Preferences, identity and competition: Predicting tie strength from demographic data. Management Science, 51(9), 1374–1383.
Reinartz, W., Krafft, M., & Hoyer, W. D. (2004). The customer relationship management process: Its measurement and impact on performance. Journal of Marketing Research, 41(3), 293–305.
Reingen, P. H., & Kernan, J. B. (1986). Analysis of referral networks in marketing: Methods and illustration. Journal of Marketing Research, 23(4), 370–378.
Reingen, P. H., Foster, B. L., Brown, J. J., & Seidman, S. B. (1984). Brand congruence in interpersonal relations: A social network analysis. Journal of Consumer Research, 11(3), 771–783.
Schmittlein, D. C., Morrison, D. G., & Colombo, R. (1987). Counting your customers: Who are they and what will they do next? Management Science, 33(1), 1–24.
Singh, J. (2005). Collaborative networks as determinants of knowledge diffusion patterns. Management Science, 51(5), 756–770.
Trusov, M., Bucklin, R. E., & Pauwels, K. (2009). Effects of word-of-mouth versus traditional marketing: Findings from an Internet social networking site. Journal of Marketing, 73(5), 90–102.
van den Bulte, C., & Moenaert, R. K. (1998). The effects of R&D team co-location on communication patterns among R&D, marketing and manufacturing. Management Science, 44(11, Part 2 of 2), S1–S18.
Volkov, M., Harker, D., & Harker, M. (2005). Who’s complaining? Using MOSAIC to identify the profile of complainants. Marketing Intelligence & Planning, 23(3), 296–312.
Author information
Authors and Affiliations
Corresponding author
Additional information
The author thanks Roger Bivand, Professor of Economics at the Norwegian School of Economics and Business Administration and developer of the spdep package, for his helpful comments during data analysis.
Appendix: Sampling process
Appendix: Sampling process
1.1 Sample A
We started the creation of sample A by taking a random sample from the customer database of the mobile phone provider. Since every customer can be uniquely identified by his/her mobile phone number, we generated a list of 150,000 random numbers and matched it to the customer database.Footnote 8 This resulted in a random sample of 363 customers. For each of these 363 customers, we downloaded information about all outgoing calls made (phone number called and duration of call) over a 3-month time period (March 1 to May 31). We then calculated the total number of minutes any number had been called and expressed it as a percentage of total call duration. Any number which represented at least 1% of total call duration was subsequently considered as a potential friend of the calling customer.Footnote 9 We then matched this list of mobile phone numbers back to the customer database, resulting in 747 customers of the mobile phone provider that could be considered as friends of at least one of the initial 363 customers. Following the same procedure again, we subsequently identified another 2,639 customers (either friends of the initial 363 and/or the 747 customers) and 6,966 customers (either friends of the initial 363 and/or 747 and/or 2,639 customers). In the resulting list of 10,715 customers (363 + 747 + 2,639 + 6,966), 2,710 customers were deleted due to double counting and 292 because they had been acquired after March 1 and, hence, only had incomplete call history information. Out of the remaining 7,713 customers, 7,055 could be matched to a second database containing revenue information and, out of those, 6,681 to a third one containing postcode data. This resulted in a final network consisting of 6,681 actors, who were linked by 19,885 call relationships or ties, with a density of 0.0005628. For our analysis, we transformed all directed relationships into undirected ones (i.e., arcs into edges) to obtain a symmetrical adjacency matrix as the underlying event (mobile phone calls) is by nature a reciprocal relationship.
1.2 Sample B
Similar to the approach taken for sample A, the creation of sample B started by matching a set of random numbers (1.25 million) to the customer database of the mobile phone provider, resulting in a random sample of 4,163 customers. Deleting 424 customers that had been acquired after August 1 led to 3,739 customers, for which we downloaded information about all incoming and outgoing calls made (phone number and duration of call) over a 2-month time period (August 1 to September 30). This resulted in 12,939 customers of the mobile phone provider that called at least one of the initial 3,739 customers and 12,853 that were called by at least one of them. In the resulting list of 29,531 customers (3,739 + 12,939 + 12,853), 9,689 customers were deleted due to double counting. Out of the remaining 19,842 customers, 19,826 could be matched to a second database containing revenue information and, out of those, 19,668 to a third one containing postcode data. This resulted in a final network consisting of 19,668 actors, who were linked by 25,799 call relationships or ties, with a density of 0.0000851. As above, we transformed all directed relationships into undirected ones (i.e., arcs into edges) to obtain a symmetrical adjacency matrix as the underlying event (mobile phone calls) is by nature a reciprocal relationship.
Rights and permissions
About this article
Cite this article
Haenlein, M. A social network analysis of customer-level revenue distribution. Mark Lett 22, 15–29 (2011). https://doi.org/10.1007/s11002-009-9099-9
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11002-009-9099-9