Phillip Alday

I believe #424 addressed this.

I think the relevant terms in general statistics are "prediction intervals" instead of "confidence intervals" (which GLM already provides). Returning a proper distribution isn't exactly a thing in the frequentist...

@pdeffebach That threshold stems from the days when that was the threshold in `LinearAlgebra` for the QR and Cholesky decompositions (cf [this discussion](https://github.com/JuliaStats/MixedModels.jl/pull/324#issuecomment-634270573)). Changing the tolerance won't always fix the...

All that said, I think we should probably - expose the tolerances - make it easier to choose the QR or Cholesky routes - add to the documentation why having...

@pdeffebach The numerical issues won't go away. Poorly conditioned models will often have misleading standard errors and other pathologies. I'm pretty strongly opposed to making things "too easy" because there...

> Is there some other heuristic we can use to determine rank deficiency outside of Cholesky? QR with Pivot, already discussed above. Singular Value Decomposition, but this is **expensive**. We...

So something like ```julia r2vals = [(-1e-5 < r

A temporary kludge would be dropping missings from the resultant `X` in the `lm(::FormulaTerm,...)` methods before `LinearModel(X, ....)` is called. That's something that should be a straightforward PR against GLM.jl.

It's unlikely that this would be added here, but there is related functionality in the Effects package: https://beacon-biosignals.github.io/Effects.jl/stable/emmeans/ It computes estimated marginal means, doesn't (yet) do tests of linear combinations...

I've been tinkering with something similar for MixedModels.jl in the form of a `MixedModelSummary` over in [MixedModelsSerialization.jl](https://github.com/JuliaMixedModels/MixedModelsSerialization.jl). We could probably define a sensible `RegressionModelSummary` interface for stored coefficients, (optionally) associated...