Statistics for Massive Datasets
Network Reliability
Communications networks are highly reliable and constantly monitored,
but sometimes call completion rates degrade abruptly. The
impact on the quality of
service depends on the severity and duration of the degradation.
Rick Becker,
Linda Clark and
Diane Lambert
have developed a new nonparametric method for estimating event rates
as a function of severity and duration.
The method provides raw estimates and
confidence intervals for both event rates
for single networks and ratios of event rates for pairs
of networks. The method
falls into an emerging area of statistical methodology that
is concerned with both statistical criteria and fast computations for
enormous databases.
Measures of Customers' Needs, Usage, and Satisfaction
Information on individual customer usage of communications services
and products is routinely kept in internal company data bases.
In addition, special surveys are often performed to measure the
satisfaction of specific customers with various aspects of their
services and purchases, or to gauge their present and future
communications needs. A continuing area of problems involves
methods for collecting and analyzing such data in order to best
answer important business questions. For example, how can detailed
information about communications needs collected on a sample of
customers be used along with internally available usage data in
order to segment the (huge) population of residence communications
customers into groups of customers with similar underlying needs?
How can customer satisfaction with various aspects of their purchases
be used to track company performance over time and suggest future
resource allocations? ``Data base mining and marketing'' to explore
such questions involves longitudinal and cross-sectional issues
and other aspects of statistical modeling and study design.
Andreas Buja,
Lorraine Denby,
Jim Landwehr
and others have been
exploring such problems in the context of specific studies with
different parts of the company.