Igor Holas posted on Monday, January 25, 2010 - 2:54 pm
I have complex survey data with censored varaibles (mostly outcomes) on which I plan on performing structural modeling.
My problem is that I have both nested data (children in schools) and need to bootstrap my standard errors (my residuals reliably fail the assumption tests).
Accounting for the nesting dramatically improved model fit.
I have ruled out recoding the censored variables (in many cases only dichotomizing solves the issue) and also transofrmations (because of interpretability issues). Bootstrapping my SE's seems like the best option.
Unfortunately, CLUSTER statement needs TYPE=COMPLEX or TYPE=TWOLEVEL, while BOOTSTRAP cannot work with those two options.
Can you please suggest a workaround?
Could I declare my [y1] as a CENSORED variable and tun TOBIT? would that approximate the BOOTSTRAP results (I was not clear on TOBIT from the documentation)