I want to know if it is possible to make predictions using a new dataset for the gamlssNP object. Using predict() directly results in an error, And I haven't seen any examples of using the predict function for mixed distribution objects(gamlssmx or gamlssNP) in the textbooks you recommended(Chapter 7 or Chapter10). I'm not sure if using predictions for mixed distribution model is reasonable mathematically or statistically.
During debugging, I found significant differences between the output of the gamlssNP object and typical gamlss object, and this difference was not considered at all in the predict.gamlss function.
For example, in predict(), fitting is performed by obtaining the formula of the fitted model object, while in the gamlssNP fitting process, modifications are made to the formula itself, such as adding the MASS factor in mu.formula. However, this MASS factor is not present on the new dataset (as it was generated during the model fitting process)
Therefore, in this situation, I am not very clear whether this mixed distribution model can be used for predicting and drawing centile curves, and whether the gamlssNP is feasible in applications such as centile estimation.
If this mixed distribution model is consistent with the typical gamlss in subsequent applications, I am willing to further share more details about debugging to help you. Looking forward to reply. Thx very much.
I want to know if it is possible to make predictions using a new dataset for the gamlssNP object. Using predict() directly results in an error, And I haven't seen any examples of using the predict function for mixed distribution objects(gamlssmx or gamlssNP) in the textbooks you recommended(Chapter 7 or Chapter10). I'm not sure if using predictions for mixed distribution model is reasonable mathematically or statistically.
During debugging, I found significant differences between the output of the gamlssNP object and typical gamlss object, and this difference was not considered at all in the predict.gamlss function.
For example, in predict(), fitting is performed by obtaining the formula of the fitted model object, while in the gamlssNP fitting process, modifications are made to the formula itself, such as adding the MASS factor in mu.formula. However, this MASS factor is not present on the new dataset (as it was generated during the model fitting process)
Therefore, in this situation, I am not very clear whether this mixed distribution model can be used for predicting and drawing centile curves, and whether the gamlssNP is feasible in applications such as centile estimation.
If this mixed distribution model is consistent with the typical gamlss in subsequent applications, I am willing to further share more details about debugging to help you. Looking forward to reply. Thx very much.