Friday, September 24, 2004
Here is my understanding about permutation test, please correct me if I am wrong. Do 999 resampling without replacement, plug in the GLRT in that unusual window(one subset only) , sort and find the position of lamda.
Yes, that's basically correct. Note you do not have to use the GRLT if you think you have a better test statistic. Also, the "sampling with replacement" has to keep the total number of injury accidents constant. Really what you are doing is randomly permuting the labels (injury or non-injury) on the accidents. D.
Yes, that's basically correct. Note you do not have to use the GRLT if you think you have a better test statistic. Also, the "sampling with replacement" has to keep the total number of injury accidents constant. Really what you are doing is randomly permuting the labels (injury or non-injury) on the accidents. D.
