SE Calculations for as.lmitt vs. lmitt

[jwasserman2]

1. DirectAdjusted objects should have a slot indicating whether it was generated using lmitt or as.lmitt.
2. When calling vcovDA on a DirectAdjusted object generated using as.lmitt, return the full covariance matrix if model-based SE's are called for, otherwise error.

[jwasserman2]

• DirectAdjusted objects have a slot indicating whether they were generated using lmitt or as.lmitt
• vcovDA called on a DirectAdjusted object generated using as.lmitt errors when design-based SE's are called for
• vcovDA called on a DirectAdjusted object generated using as.lmitt returns the full covariance matrix when model-based SE's are called for

[josherrickson]

Slot name is design_based_eligible (open to alternatives). Logical FALSE if created via as.lmitt or lmitt.lm; TRUE if created via lmitt.formula.

[benthestatistician]

How about lmitt_fit for the slot name? Maybe less potential for confusion than with design_based_eligible, which may not be evocative when our focus is elsewhere than the design based/model based distinction. I'm open to alternatives too.

[josherrickson]

That was actually the slot name I first used, but I thought a logical would be more straightforward to use in the future (rather than always having to check if (object@lmitt.fit == "lmitt").

[josherrickson]

I guess it could be lmitt_fit = TRUE in lmitt() and lmitt_fit = FALSE in as.lmitt()? Still feels a bit wonky. internal_fit?

edit: fit_with_lmitt? lmitt_called? direct_lmitt?

[benthestatistician]

lmitt_fitted?

[jwasserman2]

@xinhew0708 I think you can address the remaining two check-boxes on this issue when you create your design-based variance estimation function. Or you can let me know and I can push a commit to your branch addressing them

[benthestatistician]

(I would have thought that the second of those to-do's would fall more naturally in your court, @jwasserman2. I.e.

• vcovDA called on a DirectAdjusted object generated using as.lmitt returns the full covariance matrix when model-based SE's are called for

Of course may be missing something. Or lots of things!)

[jwasserman2]

@benthestatistician the third checkbox is already implemented, so it's more of an issue of not breaking that when addressing the second checkbox. Should have made that clearer

[xinhew0708]

Implemented checkbox 2 in function .vcov_DB0 in the design branch.