|
SPEAKER: |
|
|
TITLE: |
"Selection of Variables in Forward Stepwise Logistic Regression" |
|
DAY: |
Wednesday, March 4, 1998 |
|
TIME: |
3:30 p.m. [Coffee in BSB-202 at 3:00 p.m.] |
|
PLACE: |
BSB-108 |
The logistic regression model is used in medical research and in epidemiological studies; however, one often wishes to reduce the number of variables in such a model. The forward stepwise algorithm for a regression model involves at each step the selection of a variable for entry into the model, and the determination of whether enough variables are in the model so that the algorithm may be stopped. Differences between the use of certain selection criteria and stopping criteria are affected by factors such as the number of explanatory variables available, the degree of correlation between these variables and the sample size. This work extends to the logistic regression model methodology developed by Constanze and Afifi (1979) for the multiple regression model and adds a multivariate binary model due to Bahadur (1961). With the logistic model one seeks to minimize the error in predicting the binary outcome. Two sets of results are given: (1) the level of significance when the stopping criterion is the usual chi-square test, and (2) a comparison of the order of the variables selected by different selection criteria
Dr. John Koval received his Bachelor's and Master's degree from the University of Waterloo, his M.Phil. from Imperial College, University of London, and his Ph.D. from the University of Western Ontario under Allan Donner. He spent time as a post doctoral fellow at Waterloo, was a faculty member in three different departments at Western, and is now Associate Professor of Biostatistics at Western. His research interests are in intracluster correlation, logistic regression models and measures of agreement.
The following articles have been provided by Dr Koval to be used as background for his talk. They are on reserve at Thode Library (STATS 770: Statistics Seminar).
[1] Constanze. M.C. and Afifi, A.A. (1979). "Comparison of Stopping Rules in Forward Stepwise Discriminant Analysis," JOURNAL OF THE AMERICAN STATISTICAL ASSOCIATION 74, pp. 177-84.
[2] Kennedy, W.J. and Bancroft, T.A. (1971). "Model Building for Prediction in Regression Based Upon Repeated Significance Tests," ANNALS OF MATHEMATICAL STATISTICS 42, pp. 1273-84.
[3] Lee, K.I. and Koval, J.J. (1997). "Determination of the Best Significance Level in Forward Stepwise Logistic Regression," submitted to COMMUNICATIONS IN STATISTICS - SIMULATION AND COMPUTATION.