the model in terms of predicting the data; the lower
its value, the better is the model. Instead of detailing
the coefficients, we discuss the two main derived
expressions. (We detail the formulae in Appendix 1.)
First, we express the purchase rate in terms of
weekly transactions. Second, we compute an average
inactivity date. We obtain the results in Table II.
The quality of the estimation based on the BIC is
very specific: It is poor for the traditional transactions
(especially consumers who only buy traditional coffee),
which likely reflects their high degree of heterogeneity.
Considering the various types of traditional
coffees available as one unique product is clearly a limit
for the accuracy of our analysis. The purchase rate
expected represents the average number of coffee
transactions per week and per customer. A transaction
might include several packages (i.e., on average, 1.3
packages per transaction). Similarly, the inactivity date
expected is the average ‘‘lifespan’’ of a customer,
which reflects when the household stops purchasing
coffee or FT coffee for any reason.