Data Mining and Statistical Modeling

The Measurement Group is expert at developing empirical data models from data typically collected by healthcare and other human service organizations. Some examples of data mining and statistical modeling efforts are highlighted because they illustrate major techniques or data findings. The Measurement Group develops appropriate results models using a variety of methods ranging from data mining methods for large data warehouses to special statistical models optimized for very small samples. The links below go to representative applications of data mining and statistical modeling methods found on this web site. To find out more about the capabilities of The Measurement Group in data mining and statistical modeling, see our flyers and our How To pages.

Each of the examples below was chosen from one of the TMG Knowledge Bases or from a TMG report. When the link takes you to a "Knowledge Item," click on the buttons labeled "Extended Results" and "Additional Statistics" to see figures and technical output.

We selected the examples featured here because they may be of general interest. These examples tend to be in health-related applications because much of our public-sector work is in this area. We also do data modeling in the areas of education, marketing, test development, and public planning. Note that there are also links to two large index pages, each with literally dozens of links to specific examples for those who would like to see more examples covering many areas.

Developing a Decision Tree for Predicting Events: Quality of Life in HIV/AIDS Patient
CHAID (chi-squared automatic interaction detection) is a commonly used procedure for developing decision trees. Exhaustive CHAID is used in this example to model the factors that contribute to feelings of quality of life among clients of national demonstration programs on innovative HIV/AIDS services.

Developing a Decision Tree for Predicting Events: Patient Satisfaction with Managed Care
This example uses CHAID to study patient satisfaction among clients in innovative medical treatment programs. Subtypes of individuals who feel relatively more and less satisfied with their services are identified.

Developing a Decision Tree for Predicting Events: Determination of Which HIV/AIDS Patients Have the Greatest Gain in Quality of Life as a Result of Treatment
CHAID methods are extended here to studying individual change and how change relates to a number of patient service needs and vulnerabilities. In this example, major behavioral and demographic factors are used to model change in self-perceived quality of life over time while participating in an innovative treatment program.

Other Examples of Developing Decision Trees to Predict Events and Levels of Functioning
Link to a large index (continuously updated) of applications of decision tree modeling to develop ways of splitting groups of individuals into types so as to best predict behaviors, attitudes, and intentions.

Predicting and Explaining the Duration of Time Until an Event Occurs: Retention in Treatment
This example shows an application of methods of event history ("survival") analysis to the question of how long clients will stay in a treatment program. The methods illustrated are ones called Kaplan-Meier survival analysis and Cox proportional hazards regression.

Predicting and Explaining the Duration of Time Until an Event Occurs: Time from Enrollment Until A Service is Received
Event history analysis is used to study how much time elapses from enrollment to receiving services in community-based programs for HIV/AIDS. Cox regression methods are illustrated.

Predicting and Explaining the Duration of Time Until an Event Occurs: How Long is Quality of Life Maintained at Baseline Levels for Psychosocial, Comprehensive Healthcare, and Managed Care HIV/AIDS Patients?
Event history or survival analysis is used to study how long a patient can be maintained at baseline or a higher level of quality of life by three types of service providers. Baseline differences in functioning are controlled. Cox regression methods are illustrated.

Other Examples of Predicting the Time Until an Event Occurs
Link to a large index (continuously updated) of applications of modeling to determine the time that elapses until an event occurs.

Clustering Individuals: Behaviors of HIV/AIDS Patients
Cluster analysis is a method determining the major groupings of individuals. In this example, AIDS patients are clustered together by their behavioral characteristics so that homogeneous groups can be delineated and treatment outcomes can be studied for each group.

Clustering Individuals: Patterns of Quality of Life among HIV/AIDS Patients
This example of cluster analysis shows types of HIV/AIDS patients formed based on their patterns of quality of life.

Confirmatory Modeling of Respondent Behaviors
This example shows an application of methods of confirmatory factor analysis. Confirmatory factor analysis is used to test whether models that predict the interrelationships of variables are consistent with observed behavior. This example tests the plausibility of models which explain how indices of HIV-risk interrelate in adolescents and young adults.

Confirmatory Modeling of Service Utilization
Confirmatory factor analysis, using hierarchical constructs, is used to develop a model for service utilization by youth entering community services. The model developed is plausible for data collected in 10 national demonstration projects.

Predicting and Explaining Dichotomous Response Categories
Logistic regression methods are used to predict qualitative variables which can be coded dichotomously (such as, yes versus no, happened versus did not happen, etc.). This example shows the prediction of whether a youth has a history of substance abuse from other information about behavior.

Predicting and Explaining Multinomial Response Categories
Multinomial logistic regression methods are used to predict qualitative variables which are coded into more than two categories. This example shows the prediction of different kinds of practice change that are made by health care professionals after attending training.

Related Information:

The Measurement Group shows How to...
The Measurement Group Evaluation & Research Tools
The Measurement Group Glossary Index
Using Your Module Data: An Intermediate-Advanced Training Session Using SPSS For Windows