Talk to the Veterans Crisis Line now
U.S. flag
An official website of the United States government

Health Services Research & Development

Go to the ORD website
Go to the QUERI website

2005 HSR&D National Meeting Abstract

2008 — Classification and Regression Trees: An Important Statistical Tool for Health Researchers

Author List:
Glickman ME

Workshop Objectives:
Least-squares regression and logistic regression are among the most common statistical tools applied to health services data to assess relationships between predictor variables and a dependent variable. However, blind use of these regression techniques can result in misleading conclusions. For example, if some predictor variables are not transformed appropriately, they may be determined to be not statistically significant when they are important variables. Furthermore, potential predictors may be important in a regression only when interacted with other variables, and such interactions are often difficult to know a priori. The use of classification and regression trees (CART) as an exploratory tool, or even as an alternative to conventional regression techniques, addresses some of these difficulties. CART, developed by Breiman et al. (1984), is a flexible method based on recursive partitioning of the data to model a continuous or categorical dependent variable from a set of predictors. CART is becoming a more accepted tool in health services applications, so it is worthwhile for researchers to be aware of its use, as well as its benefits and limitations compared to conventional regression techniques. The objectives of this workshop are for participants to (1) learn the fundamentals of CART, (2) be able to interpret the results of CART applied to health services regression data, and (3) understand the relationship between CART and conventional regression methods, and how CART can help to guide modeling choices in standard regression.

After explaining the general principles of CART, and highlighting them through simple examples, the application of CART will be demonstrated on several health services data sets. Applications will be presented from the output of S-Plus statistical software. Audience participants will be provided handouts that both summarize workshop material and contain short self-test exercises.

Target Audience:
The workshop is intended for health service researchers who use regression methods in their work.

Assumed Audience Familiarity with Topic:
Workshop participants should have basic familiarity with regression at the level of an introductory statistics course. No prior knowledge of CART is assumed.

Questions about the HSR&D website? Email the Web Team.

Any health information on this website is strictly for informational purposes and is not intended as medical advice. It should not be used to diagnose or treat any condition.