It is usually desired behavior that the component estimates for the unmatched $D_A$ and $D_B$ are omitted if they are zero. However, this creates a problem wherever returns are expected for all components, e.g. for mi estimate when some imputations have full support and some don't.
I suggest to add an explicit option which replaces missing components with a very small value, e.g. 10e-10 and leaves the SE untouched. For mi estimate that would circumvent the omitted terms vary error (498) and result in a coefficient, but no SE. Not having a SE is fine, this should only affect edge cases where the component is essentially zero. If not, it should not be done anyhow.
It is usually desired behavior that the component estimates for the unmatched $D_A$ and $D_B$ are omitted if they are zero. However, this creates a problem wherever returns are expected for all components, e.g. for
mi estimatewhen some imputations have full support and some don't.10e-10and leaves the SE untouched. Formi estimatethat would circumvent theomitted terms varyerror (498) and result in a coefficient, but no SE. Not having a SE is fine, this should only affect edge cases where the component is essentially zero. If not, it should not be done anyhow.