American Statistical Association New York City Metropolitan Area Chapter Mailman School of Public Health Columbia University Department of Biostatistics Colloquium

Continuous variables are widely used in linear, logistic and Cox proportion hazards regression models. Usually these variables are simply entered "as is" into the model. Doing so, however, assumes a straight-line relationship with the outcome variable. Alternatively, the continuous variable may be grouped into g categories and modeled with g-1 dummy variables. Suppose, instead, that we would like to retain the full information in the measurements but not assume a straight-line relationship. This talk will review three methods for assessing scale for continuous variables in logistic regression models.

Dr. Lemeshow is Dean of the College of Public Health at The Ohio State University, and, author of four well-known textbooks in biostatistics on survival analysis, sampling of populations, logistic regression, and sample size in health studies.