Neal Fultz

@alexpghayes I haven't tried, but if `drop1` somehow works I think you can recursively call it + combine that with the workaround in the thread I linked before using `anova.lmlist`....

It *might* work, but IIRC nobdy has ever asked about `mlm` compatibility before or tried. Also do you need `anova` / `manova` multivariate tests or `aov` per-outcome tests ? BTW...

Apparently the method dispatches differently for iv_robust than lm_robust - please try using the lm method explicitly, like: ``` stats:::model.frame.lm(fit) ``` This is another edge case for #123

It's weird, someone in the other thread said get_data used to work? https://github.com/DeclareDesign/estimatr/issues/123#issuecomment-492163357 Maybe it changed?

See also https://stackoverflow.com/q/3872070/986793

> May be enough to flag in documentation that printed output is for single hypotheses and not for joint hypotheses when these are requested; (as noted) the original output is...

Multcomp has a bug, where it's ignoring the df's you've provided it - https://github.com/cran/multcomp/blob/fccb9a9b45b222fbab50ec47cf5778d3990ec9e2/R/helpers.R#L136-L139 it is hardcoded for lm and aov.

@acoppock I've encountered that at another job - the issue is that with the `Z:X` interaction term, you need to track the structural sparseness of that column. The bookkeeping gets...

Voting against hardcoding in Z_cond_prob into horvitz_thompson - this is related to discussion where declare_estimator needs to be able to access the assignment step during executing a design.

For now just bang bang it: model3