Hi,
Can I run bootstrap in NLME in such a way that the re-sampled individual data was stratified by a binary covariate(e.g. male and female) so that I won’t have a weird situation where all the re-sampled subjects come from one group and the parameter for the covariate cannot be estimated?
Tao
yes provided that you defined a covariate as categorical it should be available for stratified bootstrapping.
What if I’m writing the PML by myself, but the binary covariate that was defined as fcovariate(gender()) was not picked up by the program automatically in the bootstrap. The binary covariate was identified as a categorical variable and I can find it in the Pop Covariate Boxplots. Did I miss anything when I define the covariate, or there’s another way to run the stratified bootstrap?
Dear Tao
Do you have the project?
When you click on stratification, can you chose the categorical covariate as a strata. I can. Then is the problem with that or when running the strata ar enot icked up?
.​Best Regards and great to hear from you. We should schedule a workshop in the near future. To minimize cost we could do it after or before big conferences like ACOP, WCOP, PAGE.
Best Regards
Serge
Hi Serge,
Thank you for your reply. I will send the phx file to you shortly. But I think this might be a common situation in bootstrap. I am not sure if I missed anything when I defined the covariate, but it seems like PHX cannot use the categorical covariate when I switch the Run Mode from Simple to Bootstrap. So, I have to create a new model using builtin and add the categorical covariate first, then change Simple to Bootstrap. When I see the covariate in bootstrap, I can change builtin model to PML code. Then, I can run the bootstrap with my PML code.
Tao
Hi Tao, I just tried this several times and it seems to pick up the categorical variable(s) just fine for stratification when I bootstrap, I’m running 6.4 with both the PHX 29 and 30 patches applied. (So under Edit > Preferences > Plugins > DME > 1.4.0.890 is the build number I have ).
Are you absolutely sure you added the extra brackets when you added the categorical covariate in the textual mode? (In library and graphical mode this is done for you via the tab Covar. Type as I think you already noted.
If you are still seeing this problem please confirm the DME plugin version and send us the project to investigate.
Simon
I see. My version is a little old, 1.4.0.0
I will send you and Serge the file I am working on by email.
Tao
Hi Tao, I think I have solved (and recreated) your problem.
Your syntax was fine e.g
fcovariate(Azole())
However it seems in Phoenix 1.4 (patched or not) that there is a problem when you go straight to textual mode without first specifying in the UI a categorical parameter. (Note I think we have fixed it in QC14363 as it does not happen in the forthcoming 7.0 release)
If I copy your PML text into a model object I created where I previously defined covariates as categorical I can stratify in boot strap.
However if you start with a model object and go straight to text mode before defining covariates as categorical, then the stratify option in bootstrap will not appear.
Please see the ‘solved’ project to illustrate the workaround
Simon.