Statistical Programs |
College of Agriculture | University of Idaho |
Seminar Announcement |
"Applied Statistics in Agriculture" |
Small sample tests for comparing several logistic regression slopes to a standard
Presented By |
Dr. Nairanjana Dasgupta |
Program in Statistics Washington State University |
Tuesday, Dec. 5 3:30 P. M. Room 62 College of Agriculture |
Logistic regression is a common technique used for analyzing binary data. Commonly it arises in experiments in the life sciences. Often it becomes imperative to compare several logistic regression slopes to that of a control. The existing methods, Reiersol's procedure and Bonferroni Wald procedure, are both asymptotic techniques assuming large sample sizes. Often in real life we are faced with situations when sample sizes are small. We propose two tests based on the minimal sufficient statistic for this problem. For the first, we use pivoting techniques to find our critical values. For the second we use the non-parametric permutation test approach. We compare our pivoted test statistic to Reiersol's method and MLE based Bonferroni Wald method. Simulation studies indicate that the pivoted small statistic has he highest marginal power and controls Type I error close to the nominal level unlike its competitors. The permutation test is competitive in terms of marginal power and is also an exact test. The research is motivated by a problem in plant pathology, which we cite as our data example. |
Return to Seminar Listing. |