CVA and Its Important Role in Lucent
The Lucent Technologies business units commission surveys to measure the level of satisfaction of their customers and the level of satisfaction of competitor's customers. Survey results are used to manage the business at a fundamental level. They are a basis for decisions about

product selection

process improvement

employee compensation, particularly top management

Our Project: The Analysis of Survey Data
We are developing

general approaches

specific data analysis tools

and software

Our principal vehicle for drawing conclusions from the surveys is a statistical model for the data that describes the variation in customers' ratings as a function of 43 different measures of company performance, the companies (Lucent and its competitors), the individual respondents in the survey, and time. The model is quite complex because the variation in the ratings is complex. For example, different survey respondents use the rating scale differently; they position themselves at different places on the scale and they use up different amounts of the scale.

Visualization has also played an important role in our analyses. First, display of the data has been vital to developing a statistical model that does a good job of describing the actual variation in the data. That is, by studying the structure of the data, we have been able to develop a realistic model. Second, visualization has played an important role in conveying our results. The results are complex, but the visualization tools allow ready comprehension of the important messages in the data.

Research Tools Used in the Analysis
The complex process of sampling customers, surveying them, analyzing their responses, and then summarizing the results to convey the information to management is like a manufacturing line, requiring novel technologies and process improvement at all stages to provide valid characterizations of customer perceptions. A number of tools and systems invented in our statistics and data mining research organization play a major role in the analysis and summary:

Trellis Display

Bayesian modeling

The S system for graphics and data analysis

Trellis Display has been the chief visualization system used in the project, both for analysis and presentation. Trellis is a powerful system that enables us to understand the market position of Lucent relative to its competitors for several customer satisfaction measures, and to see how its market position changes through time.

The statistical model developed for the customer ratings is a Bayesian model, partly hierarchical, with a time series component and with capabilities to handle missing data. Bayesian Statistics is a major area of research for our organization. We work on foundations, models, computation, and data analysis.

The whole effort relies heavily on our S software system for graphics and data analysis which allows both rapid prototyping as well as technology transfer of final tools.

A Paper for the Workshop on Case Studies in Bayesian Statistics
Carnegie Mellon University, September 26-27, 1997
We have a long paper that describes the model for the CVA data, how we discovered it, and some of its implications for CVA.

In customer value analysis (CVA), a company conducts sample surveys of its customers and of its competitors' customers to determine the relative performance of the company on many attributes ranging from product quality and technology to pricing and sales support. The data discussed in this paper are from a quarterly survey run at Lucent Technologies.

We have built a Bayesian model for the data that is partly hierarchical and has a time series component. By ``model'' we mean the full specification of information that allows the computation of posterior distributions of the data --- sharp specifications such as independent errors with normal distributions and diffuse specifications such as probability distributions on parameters arising from sharp specifications. The model includes the following:

• survey respondent effects are modeled by random location and scale effects, a t-distribution for the location and a Weibull distribution for the scale;
• company effects for each attribute through time are modeled by integrated sum-difference processes;
• error effects are modeled by a normal distribution whose variance depends on the attribute; in the model, the errors are multiplied by the respondent scale effects.

Building the model and using it to form conclusions about CVA stimulated work on statistical theory, models, and methods:

• a Bayesian theory of data exploration that provides an overall guide for methods used to explore data for the purpose of making decisions about model specifications;
• an approach to modeling random location and scale effects in the presence of explanatory variables;
• a reformulation of integrated moving-average processes into integrated sum-difference models , which enhances interpretation, model building, and computation of posterior distributions;
• post-posterior modeling to combine certain specific exogenous information - information from sources outside of the data - with the information in a posterior distribution that does not incorporate the exogenous information;
• trellis display , a framework for the display of multivariable data.