[R] interpretation of the likelihood ratio test under *R* GLM

Michael Menke menke at email.arizona.edu
Wed May 27 20:04:12 CEST 2009


Can anyone tell me how the LRT is to be interpreted under the glm 
package when using glm for logistic regression with mutiple predictors?

family=binomial("logit"))
              drop1(Confidence.glm, test="Chisq")

The summary z-table suggests a direction of the effect, and notably the 
large LRT statistics are the significant ones. I am used to thinking of 
extremely small LRTs as significant (negative natural logarithms of 
LRTs). I must assume that the LRT in *R* is alternative hypothesis over 
null hypothesis, rather than the convention I learned of 
null/alternative, where a small number (negative logLR) represents 
strong evidence and zero/zed evidence is represented by an LR of 1 
(logLR = 0). Comments? Am I interpreting this correctly?

J. Michael Menke
University of Arizona



More information about the R-help mailing list