2 replies to this topic

[Shu-Pei Wu]

Hi Pharsight modellers,

 

Just curious on the bootstrap method that used in phoenix.

Base on the guide, it seems to be a "non-parametric" bootstrap?

And if it is, is there a way to do a "parametric" bootstrap too?

 

Thanks,

 

Shu-Pei

[Samer Mouksassi]

The bootstrap is non parametric in the sense that we randomly select patients with replacement and re-fit the model.

E.g you have 100 subjetcst

 

Bootstrap replicate 1 : we select randomly 100 subjets with replacement ( some subjects can appear more than one time others might not appear)

re-fit model on this data get parameter estimates

 

Bootstrap replicate 2 : we select randomly 100 subjets with replacement ( some subjects can appear more than one time others might not appear)

re-fit model on this data get parameter estimates

 

and so on each replicate will give you different set of parameters which are summarized and presented.

 

What do you mean by parameteric bootstrap can be achieved using the simulation/predictive check i.e simualte using the fitted random effects then refit each replicate ( sort the fitting process by replicate.

 

But typicaly the non-parameteric bootstrap is used for uncertainty estimation.

 

All depends on what are you using the bootstrap for.

 

Samer

[Shu-Pei Wu]

Thank you so much Samer for prompt response.

It is very clear now to me.

 

Thanks,

Shu-Pei