rework passing and handling of FE coefficients to simulate! for rank deficient models

[palday]

• add tests showing potential failure mode of old route

Did behavior change? Did you add need features? If so, please update NEWS.md

• add entry in NEWS.md
• after opening this PR, add a reference and run docs/NEWS-update.jl to update the cross-references.

Should we release your changes right away? If so, bump the version:

• I've bumped the version appropriately

visual example that makes the old behavior clear

using DataFrames
using MixedModels
using StableRNGs
using Statistics
using Test

id = lpad.(string.(1:40), 2, "0")
B = ["b0", "b1", "b2"]
C = ["c0", "c1", "c2", "c3", "c4"]

df = DataFrame(reshape(collect(Iterators.product(B, C, id)), :), [:b, :c, :id])
df[!, :y] .= 0

filter!(df) do row
    b = last(row.b)
    c = last(row.c)
    return b != c
end

m = LinearMixedModel(@formula(y ~ 1 + b * c + (1|id)), df)
β = 1:rank(m)
σ = 1
simulate!(StableRNG(628), m; β, σ)
fit!(m)
boot = parametricbootstrap(StableRNG(271828), 1000,  m);

bootci = DataFrame(shortestcovint(boot))
filter!(:group => ismissing, bootci)
select!(bootci, :names => disallowmissing => :coef, :lower, :upper)
transform!(bootci, [:lower, :upper] => ByRow(middle) => :mean)
@test all(x -> isapprox(x[1], x[2]; atol=0.1),  zip(coef(m), bootci.mean))

using CairoMakie
using MixedModelsMakie
let f = Figure()
    ax = Axis(f[1,1])
    ridgeplot!(ax, boot)
    coefplot!(ax, m; color=:red)
    f
end

Red are the Wald CIs, Black is bootstrap.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 96.99%. Comparing base (918457f) to head (ceb0600).

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #778   +/-   ##
=======================================
  Coverage   96.99%   96.99%           
=======================================
  Files          34       34           
  Lines        3360     3361    +1     
=======================================
+ Hits         3259     3260    +1     
  Misses        101      101           

Flag	Coverage Δ
current	96.93% <100.00%> (+<0.01%)	⬆️
minimum	96.84% <100.00%> (-0.06%)	⬇️
nightly	96.50% <100.00%> (+<0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[kleinschmidt]

this is the branch that handles the coef case where we need to pull out the non--0.0 values in the correct order to multiply by fullrank(m) right?

[palday]

correct.

[kleinschmidt]

I think it's worth documenting somewhere the difference between the possible kinds of inputs to this function. coef-style (all values in input model matrix order, pivoted out betas are -0.0) vs. fixef-style (only betas for full-rank columns, in pivot order)

[palday]

73d50ae