| boral | R Documentation |
This list below is written in terms of what is deemed to be most important to least important changes.
Zero truncated Poisson and zero truncated negative binomial distributions are now available, and fingers crossed they work! They are accessed by setting family = "ztpoisson" and family = "ztnegative.binomial". Please note that, as is commonly used in other regression models, the models are set up such that a log-link function connects the mean of the untruncated distribution to the linear predictor.
Creator, author, and contributor have been more formally defined now. Furthermore. lifecycle stickers have also been included for each function, courtesy of the lifecycle package although, it is all a bit “opinionated" in the way it is applied. This are most clearly visible when using boral in RStudio.
The function fitted.boral now has an additional argument linear.predictor. When set to TRUE, it allows one to return the fitted values on the linear predictor scale. When set to FALSE, which is the defaul behavior, the fitted mean values are on the response scale. Note things are slightly more complicated for the two newly implemented zero truncated distributions because, as mentioned above, the log-link function connects the mean of the untruncated distribution to the linear predictor. Therefore if linear.predictor = TRUE, then the linear predictor is returned. But if linear.predictor = FALSE, then actual fitted mean value is returned.
The default for value for mcmc.control$seed has been changed from 123 to NULL. Thus the user must now specify a seed if they wish to "seed" the MCMC fitting algorithm.
An issue pertaining to the calculation of the Dunn-Smyth residuals for Tweedie responses has been fixed. Thaks to an ongoing collobration with Adrien Chevallier to leading to this issue being rectified.