Model comparisons without data
Opened this issue · 2 comments
ast0815 commented
It would be useful to create model comparisons without data, i.e. determine whether two models are distinguishable given a detector response.
ast0815 commented
Idea:
- Create toy data sets according to model A
- Calculate p-value of model B for each toy
- Figure of merit is fraction of toys that exclude model B at desired confidence level
Atlernative idea:
- Create toy data set that maximises likelihood of model A
- Figure of merit is p-value of model B
ast0815 commented
Third idea:
- Generate toy data sets according to models A and B
- Figure of merit is Mahalanobis distance between toy data sets