Douglas Bates

@hikea Regarding the failure to produce a warning for models including random effects for `NativeLanguage` by `Subject`, the fitted model is obviously singular and I presume that the user is...

@hikea Regrettably that fit fails in the development version of `MixedModels`. I need to patch up some starting estimates ```julia julia> lexdec_glmm_gamma = fit!(glmm(@formula(rt_raw ~ 1 + Class*NativeLanguage + (1|Subject)...

The `fast=true` version of the fit uses Penalized Iteratively Reweighted Least Squares (PIRLS) to determine the conditional mode of the random effects and the conditional estimate of the fixed-effects parameter,...

@hikea You have almost convinced me that I should print warning messages about convergence to singular covariance matrices. All of GLMM fits are to singular covariance matrices for the `Subject`...

On a related issue, I have been looking at the method for `leverage` in `src/mixedmodels.jl`. As stated in the comments, this is a very inefficient way of determining these values....

The good news is that there is low-hanging fruit for speeding up evaluation of the leverage values. I'll start on that today but it may take a few days.

I think the formulation in terms of elementary operations using the blocks themselves is sufficiently straightforward to implement. Each leverage value uses only one diagonal block from `m.L[1]` in the...

Do you want to continue to use `StableRNG` now that the `Random` package for v1.7.0 has switched to `Xoshiro` as the default RNG?

I can work on some of these if that would help.

I'm wondering how far down the rabbit hole to jump with this one. (It will probably be very far if past experience is a guide.) It seems we are inconsistent...