tunity to gain valuable insights from the statistics literature on the modeling of rare events (King and Zeng 2001). 10. Recognizing the Dangers of Endogeneity It is well known that the statistical significance of a model parameter does not necessarily imply that the cor- rect variable was used and that the effects we are mea- suring are valid. We need to focus on the relevant theory, understand the threats of confounds such as endogeneity, and develop the appropriate modeling methodologies. The issue of endogeneity has received considerable atten- tion in the past decade (e.g., Wittink 2005). There is con- siderable literature in the field of new empirical industrial organization (I/O) that deals with the issue of endogene- ity (e.g., Berry, Levinsohn, and Pakes 1995). However, its proposed solutions (such as instrumental variables and VAR models discussed earlier) put additional demands on the data that are not always within reach of the CLV modeler. In some cases, CLV parameters may have to be estimated from experimental designs in which endogene- ity is eliminated. We need to understand to what extent endogeneity is a threat to our models, not only in theory, but also in practice. 11. Accounting for Network Effects Most of the research on CLV has implicitly assumed that the value of a customer is independent of other cus- tomers. In many situations, customer network effects can be strong, and ignoring them may lead to underestimat- ing CLV. Hogan, Lemon, and Libai (2003) showed that word of mouth or direct network effects can be quite substantial for online banking. Villanueva, Yoo, and Hanssens (2006) found that word-of-mouth acquisitions are twice as valuable to the firm as customer acquisitions through traditional marketing instruments. As word of mouth and buzz marketing become more and more important, we need to have a better understanding of these phenomena and how they contribute to the value of a customer over and beyond his or her purchases. In many situations, there are also strong indirect net- work effects. Consider the case of Monster.com, an employment marketplace where job seekers post their resumes and firms sign up to find potential employees. Monster provides this service free to job seekers and makes money by charging the employers or the firms. How much should Monster spend to acquire a job seeker? Traditional models of CLV cannot answer this question because job seekers do not provide any direct revenue. This indirect network effect is not limited to employment services only (e.g., Monster, Hotjobs, Craigslist) but also extends to any exchange with multiple buyers and sellers (e.g. real estate, eBay). Research in the social network theory can be very useful for exploring these issues (e.g., Newman 2003; Wasserman and Faust 2005; Watts 2004).
You've reached the end of your free preview.
Want to read all 18 pages?