In Zadchow/concurve: Computes & Plots Compatibility (Confidence), Surprisal, & Likelihood Distributions

Here are some to-dos for concurve in the future.

• More unit tests for various purposes as Alex Hayes describes in his wonderful article. He classifies them into three types, which I copy and paste below:
• "Correctness tests check whether the code calculates the quantity it is supposed to calculated."
• "Parameter recovery tests check whether the implementation can recover correct parameters in well understood scenarios."
• "Convergence tests check whether iterative fitting procedures have actually reached a solution."
• "Identification tests check whether the solution is unique, and stable under small perturbations to the data."

---

• Make curve_rev() more usable for a wide range of scenarios (beyond proportions, and normal approximations). The confidence functions they produce especially break down with continuous variables, so I have completely disabled the option for now, and will implement an approach by working backwards from the deviance and log-likelihood functions.

---

• Produce consonance/confidence functions for quantile regressions from the quantreg package
• Implementation with the nlme package, although that has already started with some experimental work I and the other developers have done with the generalized least squares method from nlme.

---

• Possibly incorporate rstan via rstantools to produce likelihoods and posteriors

If you'd like to help contribute, please see the contributing guide.

R Package Citations

citation("quantreg")
citation("nlme")
citation("rstan")
citation("rstantools")

Zadchow/concurve documentation built on Jan. 11, 2024, 4:55 a.m.