sas macro

noetsi

Fortran must die
#1
There is a sas macro for bootstrapping when running multilevel analysis with a categorical dependent variable. It is used when the number of second level units is so small that you need to run non-parametric bootstrapping. I have the macro that does this with interval data, but not categorical data.

If anyone has this, please let me know :) I do not know the name of the macro.
 

hlsmith

Less is more. Stay pure. Stay poor.
#2
Never heard of it, but bootstrapping in sas is crazy easy. Just create m-sets in Surveyselect, then run the data through your model, typically using strata option or naming the sets sequally (dataset_1,...dataset_m). Post your model code and i can help write it on the 27th if i have time.
 

noetsi

Fortran must die
#3
I can do bootstrapping. This is a specific program, of which bootstrapping is part, to run PROC GLIMMIX when the number of groups is too small to generate correct results normally. I don't have the code although I can send the code to deal with interval DV (I really need the code to deal with categorical DV).

Thanks.
 

hlsmith

Less is more. Stay pure. Stay poor.
#5
I believe with MLM data you gotta bootstrap a little differently to procure representation in groups. I saw a wicklin or SAsUser paper on it. The groups were states.
 

noetsi

Fortran must die
#6
I don't understand post 4 hlsmith.

If you know where the paper for categorical data lives please send it. :) I have one for interval data.
 

hlsmith

Less is more. Stay pure. Stay poor.
#9
Which author did you try to contact? Contact them all; and there is a 2017 paper on this - contact those supplemental authors as well. The 2006 paper has code, is it the macro before they named it or an earlier version, see below links.






[/CODE]
 

noetsi

Fortran must die
#11
I was finally able to open the link in post 10 but the appendix which has the macro is not there (I have to pay for it I assume).

The only thing that shows up in post 9 is a lot of X's...

thanks for posting it.
 

hlsmith

Less is more. Stay pure. Stay poor.
#12
If you are feeling hardcore, you can create a field of fictitious residuals, worst case just create it by hand. Then you can recreate the IML part of the code and run the macro. I can then run the synthetic data on my end using IML and confirm you wrote it correctly. Ideally if you have the data you could have me run it first so you could tweak your code while designing it.