- Bootstrap the fitted models (rather than allow the parameters to be refit, simply evaluate the likelihood of the model under each simulated dataset)
- Directly evaluate Neyman-Pearson lemma for each pairwise model comparison.
- Also analyze the bootstrapped likelihoods directly (and without refitting) for each model.
To Do
- Add a summary function for the likelihood_ratio_bootstrap.R library
- final wrapper function for infer_niches.R partitioning library
- More error handling and documentation
- Example of partitioning Anoles then bootstrapping
- Bootstrapping likelihoods directly
References on thinking about bootstrapping model comparisons
- The original Neyman-Pearson paper
- Testing Statistical Hypotheses Text by Lehmann & Romano
- Permutation, Parametric and Bootstrap Tests of Hypotheses by Good
- Thanks to Peter on these; Bibsonomy collection
ouch2ape.R function needs debugging! Works only some of the time. Seems to error on converting back from an ape2ouch conversion.
Computing
- Adding documentation and testing the new bootstrapping functions.
- New functions make up the bootstrapping likelihood ratio suite:
- BM_loglik
- my_update
- LR_bootstrap
- LR_bootstrap_all
- plot.LR_boots
The function should be easy to call on a list of ouch models.
plot(LR_bootstrap_all(model_list))Inline documentation provides more details. Functions in this library are contained in the likelihood_ratio_bootstrap.R file.