robust SE for beta about frailtyem HOT 4 OPEN

OK with strata this became even more necessary because the correlation matrix is a whole mess.

One idea is: take the correlation matrix for the betas from the Cox model with the frailty as offset. This is provided by the mcox function.
Then do the rest of the trick with adjusting this matrix for stuff.

Upsides:

• should be waaaay faster than what is currently happening.
• should work with strata() in a pretty straight forward way, since those SE don't depend on the estimated hazard.

Downside:

• I would lose the CI in the predict function for the survival functions & stuff. Is this trade-off worth it?

from frailtyem.

Idea is to apply Louis formula for the partial likelihood

from frailtyem.

e.g. for the efficient score!

from frailtyem.

OK, that looks very difficult since Z it is also on numerator and on denominator, and then how the hell do we take the expectation.

Abandoning the part with the hazard is also not an option because then it would just take the information of beta given fixed hazard.

Therefore, currently on hol.d

from frailtyem.