Abstract:
Mixed model methods have not yet become a
common tool of the business statistician. Instead, longitudinal data are
typically summarized to one row per subject in preparation for a
modeling analysis. This often takes the form of "rolling up"
transactions to the level of a customer or account.
This workshop will offer mixed model
methods as an alternative approach. Transactions will be treated as
repeated measures on a subject. The correlation between measures will be
taken into account through either estimation of the error covariance
between measures or estimation of a random effect. In addition, we will
explore the use of best linear unbiased predictions to identify under-
or over- performing (i.e. unusual or outlying) subjects (e.g. customers,
bank branches, brokers, etc). Finally, we will demonstrate the use of
random effects modeling in a marketing experiment.
During the workshop, participants will
practice the methods on real data from finance and telecommunications.
Instructor:
John Amrhein began his career as a
statistician with the US Department of Agriculture where he designed
probability sampling techniques for agri-business surveys. While at USDA,
he provided assistance to central statistics bureaus in China,
Kazakhstan, and Romania. For the past 8 years, John has been at SAS
Institute (Canada) Inc. as an instructor and consultant in banking and
finance, telecommunications, retail, pharmaceutical, health care,
academia, and government. He has developed expertise in data mining and
predictive modeling, mixed models analysis, survival analysis, and the
analysis of complex surveys. John maintains the Professional
Statistician accreditation with the Statistical Society of Canada.