Updated predict.sclr to not have named columns with the new version of tibble. fa696d9
Made linear predictor variance calculation faster in predict.sclr. fa696d9
Reparameterised the model so that all of the parameters are unconstrained. New baseline is the logit transformation of the old baseline.
Added the gradient ascent algorithm to handle cases with high baseline.
Added a warning for a possible baseline of 1.
Added the ability to check for a possible baseline of 1 with check_baseline.
Added logLik method to access likelihood from the fit object.
Added a warning message when the model is fit with no covariates.
Added sclr_ideal_data function to simulate ideal data for the model.
Made simulations in data-raw self-contained.
Added the ability to return parameter names that are more conventional (e.g. “(Intercept)” instead of “beta_0”). See conventional_names argument in ?sclr.
Made convergence stricter to avoid local maxima. Argument n_conv to sclr and sclr_fit sets the number of times the algorithm has to converge. Best set (the one with maximum likelihood) is chosen out of n_conv sets. Previously, the algorithm only converged once.
sclr_log_likelihood can now be called with a model matrix and a model response.
Minor performance optimisations.
First release.
Fits the scaled logit model using the Newton-Raphson method.
Supports the predict method for the expected value of the linear beta X part of the model.
Can look for covariate values corresponding to a particular protection level.