|
Activity Number:
|
430
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Thursday, August 12, 2004 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
General Methodology
|
| Abstract - #301395 |
|
Title:
|
Evaluation of Approximate P Values in Logistic Regression
|
|
Author(s):
|
Xiaohong Zhang*+ and Lynn R. LaMotte
|
|
Companies:
|
Iowa State University and Louisiana State University
|
|
Address:
|
Dept. of Statistics, Ames, IA, 50011,
|
|
Keywords:
|
approximate p value ; exact p value ; chi-square statistics ; logistic regression
|
|
Abstract:
|
The logistic regression model is widely used in applications where the response variable of interest takes binary values. Exact p values for testing coefficients equal to zero can be obtained from binomial probabilities. When the number of predictor variables in regression models or observations is considerably large, acquiring exact p values may cause intensive computational difficulties. In such cases, p values obtained from the chi-squared distribution provide a way of approximation. We explore the accuracy of this approximation method in this work. Fisher-Scoring method is used to simulate SAS built-in procedures to get estimated p values and test statistics. The exact p values are obtained by computing the probabilities from binomial distributions. The datasets with different numbers of predictors, different values of x, and different number of observations of each x are used as examples. Agreement of approximate p values and exact p values is good in most cases. Approximate p value is more conservative than exact p value in the overall performance. Decreasing the number of predictors orincreasing the number of observations of each x can increase the agreement.
|
