MixedModelsMakie.jl API

coefplot(xs::Union{MixedModel,MixedModelBootstrap}...; kwargs...)::Figure
coefplot!(fig::Union{GridLayoutBase.GridLayout, GridLayoutBase.GridPosition, GridLayoutBase.GridSubposition, Makie.Figure}, xs::Union{MixedModel,MixedModelBootstrap}...;
          show_legend=length(xs) > 1, legend_attributes=(;), kwargs...)
coefplot!(ax::Axis, xs::Union{MixedModel,MixedModelBootstrap}...;
          conf_level=0.95, vline_at_zero=true, show_intercept=true,
          ptype=nothing,
          scatter_attributes=(;),
          errorbars_attributes=(;),
          show_legend=length(xs) > 1,
          legend_attributes=(;),
          labels=string.(1:length(xs)),
          attributes...)

Create a coefficient plot of the requested parameters and associated confidence intervals. When multiple models are supplied, they are overlaid on the same axes for comparison; all models must share the same coefficient names.

For a MixedModelBootstrap, ptype selects which parameters to plot: :β (fixed effects, default), :σ, :ρ, or :θ (see ridgeplot for details; the ASCII aliases :sigma, :rho, :theta are also accepted). ptype is not supported for a plain MixedModel.

group restricts ptype ∈ (:σ, :ρ) to a single grouping factor, e.g. group=:subj. It is not supported for :β/:θ or for a plain MixedModel.

attributes are passed onto both scatter! and errorbars!, while scatter_attributes and errorbars_attributes are passed only onto scatter! and errorbars!, respectively. (Starting with Makie 0.21, unsupported attributes for a given plottype are no longer silently ignored, so it's necessary to separate out the attributes that are only valid for a single plot type.)

labels controls the legend entry for each model (defaults to "1", "2", ...).

show_legend controls placement of the legend. Accepted values:

• false: no legend
• true or :bottom: horizontal legend below the axis (default for multi-model figures)
• :top: horizontal legend above the axis
• :left: vertical legend to the left of the axis
• :right: vertical legend to the right of the axis
• :axis: legend embedded inside the axis via axislegend

legend_attributes is a named tuple of keyword arguments forwarded to the Makie Legend (or axislegend) constructor.

The mutating methods return the original object.

ridgeplot(xs::MixedModelBootstrap...; kwargs...)::Figure
ridgeplot!(fig::Union{GridLayoutBase.GridLayout, GridLayoutBase.GridPosition, GridLayoutBase.GridSubposition, Makie.Figure}, xs::MixedModelBootstrap...;
           show_legend=length(xs) > 1, legend_attributes=(;), kwargs...)
ridgeplot!(ax::Axis, xs::MixedModelBootstrap...;
           conf_level=0.95, vline_at_zero=true, show_intercept=true,
           ptype=:β,
           scatter_attributes=(;),
           errorbars_attributes=(;),
           band_attributes=(;),
           lines_attributes=(;),
           show_legend=length(xs) > 1,
           legend_attributes=(;),
           labels=string.(1:length(xs)),
           attributes...)

Create a ridge plot for the bootstrap samples of the requested parameters. When multiple bootstrap objects are supplied, they are overlaid on the same axes for comparison; all inputs must share the same coefficient names.

ptype selects which bootstrap parameters to plot: :β (fixed effects, default), :σ (random-effect standard deviations, including the residual), :ρ (random-effect correlations), or :θ (the unconstrained Cholesky parameterization, labeled positionally as θ01, θ02, ... since these lack a natural per-group name). The ASCII aliases :sigma, :rho, :theta are also accepted. show_intercept only applies to :β; it is a no-op for :σ/:ρ/:θ.

group restricts ptype ∈ (:σ, :ρ) to a single grouping factor, e.g. group=:subj. It is not supported for :β/:θ.

Densities are normalized so that the maximum density is always 1. For bounded parameters (:σ ≥ 0, :ρ ∈ [-1, 1], :θ per-element via lowerbd), the density curve is truncated to the parameter's valid range, since kernel smoothing can otherwise leak density past a hard boundary.

Setting histogram=true plots a histogram instead of a KDE for each row (counts normalized the same way). A histogram respects a parameter's bounds by construction, and can show a genuine spike where many bootstrap draws land on a boundary (e.g. a singular fit) — something a smoothed KDE cannot represent. bins is forwarded to StatsBase.fit(Histogram, ...; nbins=bins) when histogram=true; otherwise StatsBase's automatic bin selection is used.

The highest density interval corresponding to conf_level is marked with a bar at the bottom of each density. Setting conf_level=missing removes the markings for the highest density interval.

attributes are passed onto coefplot, band! and lines!. scatter_attributes and errorbars_attributes are passed only onto coefplot. band_attributes and lines_attributes are passed only onto band! and lines!, respectively. (Starting with Makie 0.21, unsupported attributes for a given plottype are no longer silently ignored, so it's necessary to separate out the attributes that are only valid for a single plot type.)

labels controls the legend entry for each model (defaults to "1", "2", ...).

show_legend controls placement of the legend. Accepted values:

• false: no legend
• true or :bottom: horizontal legend below the axis (default for multi-model figures)
• :top: horizontal legend above the axis
• :left: vertical legend to the left of the axis
• :right: vertical legend to the right of the axis
• :axis: legend embedded inside the axis via axislegend

legend_attributes is a named tuple of keyword arguments forwarded to the Makie Legend (or axislegend) constructor.

The mutating methods return the original object.

ridge2d(bs::MixedModelBootstrap; ptype=:β, kwargs...)
ridge2d!(f::Union{GridLayoutBase.GridLayout, GridLayoutBase.GridPosition, GridLayoutBase.GridSubposition, Makie.Figure}, bs::MixedModelBootstrap; ptype=:β)

Plot pairwise bivariate scatter plots with overlain densities for a bootstrap sample.

ptype specifies the set of parameters to examine, e.g. :β, :σ, :ρ, :θ (the ASCII aliases :sigma, :rho, :theta are also accepted).

The mutating methods return the original object.

RanefInfo

Information on random effects conditional modes/means, variances, etc.

Used for creating caterpillar plots.

Fields

• cnames::Vector{String}: Column names, i.e. predictor coefficient names

• levels::Vector: Levels of the random effect, i.e. group names

• ranef::Matrix{T} where T<:AbstractFloat: Conditional modes (or means) of the random effects

• stddev::Matrix{T} where T<:AbstractFloat: Conditional standard deviations of the random effects

ranefinfo(m::MixedModel)

Return a NamedTuple{fnames(m), NTuple{k, RanefInfo}} from model m

ranefinfo(m::LinearMixedModel, gf::Symbol, re=ranef(m))

Return a RanefInfo corresponding to the grouping variable gf in model m.

The optional re argument is a precomputed ranef(m) result, used to avoid recomputation.

ranefinfotable(ri::RanefInfo)
ranefinfotable(ris::NamedTuple)
ranefinfotable(m::MixedModel, args...; kwargs...)

Return the information in ri (or derived from m) as a column table (NamedTuple of Vectors)

The columns are

• name: name of the random effect
• level: level of the grouping factor
• cmode: conditional mode of the random effect
• cstddev: conditional standard deviation of the random effect

When called with a NamedTuple (as returned by ranefinfo(m::MixedModel)), a group column is prepended containing the grouping factor name for each row.

The MixedModel method is a convenience wrapper equivalent to ranefinfotable(ranefinfo(m, args...; kwargs...)).

caterpillar(m::MixedModel, gf::Symbol=first(fnames(m)); kwargs...)::Figure
caterpillar!(f::Union{GridLayoutBase.GridLayout, GridLayoutBase.GridPosition, GridLayoutBase.GridSubposition, Makie.Figure}, m::MixedModel,
             gf::Symbol=first(fnames(m)); kwargs...)
caterpillar!(f::Union{GridLayoutBase.GridLayout, GridLayoutBase.GridPosition, GridLayoutBase.GridSubposition, Makie.Figure}, r::RanefInfo;
            orderby=1, cols::Union{Nothing,AbstractVector}=nothing,
            dotcolor=(:red, 0.2), barcolor=:black,
            vline_at_zero::Bool=false)

Create a "caterpillar plot" of the random-effects conditional means and prediction intervals.

A "caterpillar plot" is a horizontal error-bar plot of conditional means and standard deviations of the random effects.

When passing a MixedModel, gf specifies which grouping variable is displayed. Alternatively, ranefinfo may be used to construct the RanefInfo object directly. Constructing RanefInfo directly can be used to avoid re-computing the conditional variances.

The order of the levels on the vertical axes is increasing orderby column of r.ranef, usually the (Intercept) random effects. orderby can be an integer column index, a column name (as a Symbol or String), or nothing to disable sorting. Setting orderby=nothing returns the levels in the order they are stored in.

The display can be restricted to a subset of random effects associated with a grouping variable by specifying cols, either by indices or term names.

The mutating methods return the original object.

qqcaterpillar(m::MixedModel, gf::Symbol=first(fnames(m)); kwargs...)::Figure
qqcaterpillar!(f::Union{GridLayoutBase.GridLayout, GridLayoutBase.GridPosition, GridLayoutBase.GridSubposition, Makie.Figure}, m::MixedModel,
               gf::Symbol=first(fnames(m)); kwargs...)
qqcaterpillar!(f::Union{GridLayoutBase.GridLayout, GridLayoutBase.GridPosition, GridLayoutBase.GridSubposition, Makie.Figure}, r::RanefInfo;
               cols::Union{Nothing,AbstractVector}=nothing,
               dotcolor=(:red, 0.2), barcolor=:black,
               vline_at_zero::Bool=false)

Create a caterpillar plot with the vertical axis on the Normal() quantile scale.

When passing a MixedModel, gf specifies which grouping variable is displayed. Alternatively, ranefinfo may be used to construct the RanefInfo object directly. Constructing RanefInfo directly can be used to avoid re-computing the conditional variances.

The display can be restricted to a subset of random effects associated with a grouping variable by specifying cols, either by indices or term names.

The order of the levels on the vertical axes is increasing orderby column of r.ranef, usually the (Intercept) random effects. Setting orderby=nothing will disable sorting, i.e. return the levels in the order they are stored in.

The mutating methods return the original object.

ShrinkageInfo

Information on random effects compared to pseudo-OLS estimates.

Used for creating shrinkage caterpillar plots.

Fields

• cnames::Vector{String}: Column names, i.e. predictor coefficient names

• levels::Vector: Levels of the random effect, i.e. group names

• blups::Matrix{T} where T<:AbstractFloat: The conditional modes of the random effects

• blimps::Matrix{T} where T<:AbstractFloat: The 'reference' mode of the random effect, corresponding to the conditional mode evaluated at the reference value of θ. Typically, this approximates the value you would get without any shrinkage, e.g. from classical within-groups (non-mixed) regression

• λ::Matrix{T} where T<:AbstractFloat: The (lower-triangular) relative covariance factor of the random effects for this grouping factor, such that the covariance matrix is proportional to λλ'. Shared across all levels. Used for constructing correlation ellipses in shrinkage plots.

shrinkageinfo(m::MixedModel, θref::Vector{<:AbstractFloat}=_ref_theta(m))

Return a NamedTuple{fnames(m), NTuple{k, ShrinkageInfo}} from model m

The optional θref argument is the θ vector at which to compute the reference (unshrunken) estimates; it defaults to _ref_theta(m).

shrinkageinfo(m::MixedModel, gf::Symbol, θref::Vector{<:AbstractFloat}=_ref_theta(m))

Return a ShrinkageInfo corresponding to the grouping variable gf in model m.

The optional θref argument is the θ vector at which to compute the reference (unshrunken) estimates; it defaults to _ref_theta(m).

shrinkageinfotable(si::ShrinkageInfo)
shrinkageinfotable(sis::NamedTuple)
shrinkageinfotable(m::MixedModel, args...; kwargs...)

Return the information in si (or derived from m) as a column table (NamedTuple of Vectors)

The columns are

• name: name of the random effect
• level: level of the grouping factor
• cmode: conditional mode of the random effect
• rmode: reference mode of the random effect, corresponding to the conditional mode evaluated at the reference value of θ. Typically, this approximates the value you would get without any shrinkage, e.g. from classical within-groups (non-mixed) regression

When called with a NamedTuple (as returned by shrinkageinfo(m::MixedModel)), a group column is prepended containing the grouping factor name for each row.

The MixedModel method is a convenience wrapper equivalent to shrinkageinfotable(shrinkageinfo(m, args...; kwargs...)).

No documentation found for public binding MixedModelsMakie.shrinkagedot.

MixedModelsMakie.shrinkagedot is a Function.

# 2 methods for generic function "shrinkagedot" from MixedModelsMakie:
 [1] shrinkagedot(m::MixedModels.MixedModel, gf::Symbol; kwargs...)
     @ ~/work/MixedModelsMakie.jl/MixedModelsMakie.jl/src/shrinkage.jl:419
 [2] shrinkagedot(m::MixedModels.MixedModel; ...)
     @ ~/work/MixedModelsMakie.jl/MixedModelsMakie.jl/src/shrinkage.jl:419

shrinkageplot(m::MixedModel, gf::Symbol=first(fnames(m)), θref=_ref_theta(m); kwargs...)::Figure
shrinkageplot!(f::Union{GridLayoutBase.GridLayout, GridLayoutBase.GridPosition, GridLayoutBase.GridSubposition, Makie.Figure}, m::MixedModel,
               gf::Symbol=first(fnames(m)), θref=_ref_theta(m); kwargs...)
shrinkageplot!(f::Union{GridLayoutBase.GridLayout, GridLayoutBase.GridPosition, GridLayoutBase.GridSubposition, Makie.Figure}, si::ShrinkageInfo;
               ellipse=false, ellipse_scale=1, n_ellipse=5,
               cols::Union{Nothing,AbstractVector}=nothing,
               shrunk_dotcolor=(:blue, 0.25), ref_dotcolor=(:red, 0.25),
               ellipse_color=:green, ellipse_linestyle=:dash,
               labels::Union{Bool,Symbol,AbstractVector}=false,
               labelcolor=:black, labelsize=10, n_labels::Integer=5)

Create a scatter-plot matrix of the conditional means, b, of the random effects for grouping factor gf.

Two sets of conditional means are plotted: those at the estimated parameter values and those at θref. The default θref results in Λ being a very large multiple of the identity. The corresponding conditional means can be regarded as unpenalized.

The display can be restricted to a subset of random effects associated with a grouping variable by specifying cols, either by indices or term names.

The reference (unshrunk) points can be labeled with the levels of gf by passing labels=true (label every level), a vector of level names/indices (label only that subset), or labels=:auto (automatically pick the n_labels "most interesting" levels). labelcolor and labelsize control the appearance of the labels; n_labels only applies to labels=:auto and defaults to a small number so the plot doesn't get overcrowded.

Correlation ellipses can be added with ellipse=true, with the number of ellipses controlled by n_ellipse. The ellipses are equally spaced between the outer ellipse and the origin (center). The scaling of the ellipses can be adjusted with the multiplicative ellipse_scale. If you are unable to see the ellipses, try increasing ellipse_scale.

The mutating methods return the original object.

See also shrinkagedot.

upsetplot(m::MixedModel,
          gf::Union{Symbol,Nothing}=first(fnames(m)); kwargs...)::Figure
upsetplot!(f::Union{GridLayoutBase.GridLayout, GridLayoutBase.GridPosition, GridLayoutBase.GridSubposition, Makie.Figure}, m::MixedModel,
           gf::Union{Symbol,Nothing}=first(fnames(m));
           sortby::Symbol=:count,
           show_empty::Bool=true,
           orientation::Symbol=:horizontal,
           show_setsize::Bool=true,
           intersection_pos::Symbol=...,
           setsize_pos::Symbol=...,
           filled_color=:black,
           empty_color=:lightgray,
           bar_color=:steelblue,
           union_color=:gray30,
           dot_size=12)

Create an UpSet plot showing which levels of grouping factor gf appear in which categorical fixed-effect conditions.

Pass gf=nothing to count observations instead of grouping-factor levels.

Predictor names, levels, and per-observation values are recovered from the model's formula, design matrix, and contrast coding — no original data frame is needed.

Layout keywords:

• orientation: :horizontal (default — combinations as columns, sets as rows) or :vertical (combinations as rows, sets as columns).
• show_setsize: show the set-size bar chart (true by default).
• intersection_pos: where the intersection-size bars go relative to the incidence matrix — :top/:bottom (horizontal) or :left/:right (vertical).
• setsize_pos: where the set-size bars go — :left/:right (horizontal) or :top/:bottom (vertical).

Marginal cells (which collapse one predictor, leaving all of its levels active) get an "or" bracket drawn around that predictor's sibling dots, since lighting up every level of a predictor represents "not constrained on this predictor" (a union), not an intersection like the solid line connecting the rest of the dots.

The mutating method returns the original object.

upsetplot(data; cols=All(), gf::Union{Symbol,Nothing}=nothing, kwargs...)::Figure
upsetplot!(f::Union{GridLayoutBase.GridLayout, GridLayoutBase.GridPosition, GridLayoutBase.GridSubposition, Makie.Figure}, data;
           cols=All(),
           gf::Union{Symbol,Nothing}=nothing,
           sortby::Symbol=:count,
           show_empty::Bool=true,
           orientation::Symbol=:horizontal,
           show_setsize::Bool=true,
           intersection_pos::Symbol=...,
           setsize_pos::Symbol=...,
           filled_color=:black,
           empty_color=:lightgray,
           bar_color=:steelblue,
           union_color=:gray30,
           dot_size=12)

Create an UpSet plot directly from a Tables.jl-compatible table.

Non-numeric columns (optionally restricted by cols) become the sets. Pass gf=:col to count unique values of that column per cell instead of observations.

See upsetplot!(::Indexable, ::MixedModel) for layout keyword details.

The mutating method returns the original object.

upsettable(m::MixedModel, gf::Union{Symbol,Nothing}=first(fnames(m)))

Return the incidence table underlying upsetplot!: one row per combination cell (every full factorial cell of the categorical fixed-effect predictors, plus marginal cells that collapse a single predictor), with:

• cell: a label for the combination
• degree: the number of predictors involved (one fewer for marginal cells)
• count: the number of observations (or grouping-factor levels, if gf is given) falling in the cell
• one Bool column per set (an individual condition level, e.g. "gender: M") indicating whether that set is active in the cell

Pass gf=nothing to count observations instead of grouping-factor levels, matching upsetplot!.

upsettable(data; cols=All(), gf::Union{Symbol,Nothing}=nothing)

Return the incidence table underlying upsetplot!, built directly from a Tables.jl-compatible table. See upsetplot! for the meaning of cols and gf.

nestingplot(m::MixedModel, gfs::Union{Symbol,AbstractString,Integer}...;
            colormap=:Blues, fontsize::Real=20, swap_triangles::Bool=false)::Figure
nestingplot!(f::Union{GridLayoutBase.GridLayout, GridLayoutBase.GridPosition, GridLayoutBase.GridSubposition, Makie.Figure}, m::MixedModel,
             gfs::Union{Symbol,AbstractString,Integer}...;
             colormap=:Blues, fontsize::Real=20, swap_triangles::Bool=false)

Create a nesting/crossing matrix for the grouping factors of m.

Each distinct grouping factor (e.g. subj, item) becomes one row/column of a matrix, laid out like a correlation-matrix plot:

• diagonal: factor name and number of levels
• lower triangle (upper if swap_triangles=true): a heatmap of the co-occurrence contingency table between the row and column factor. Rows and columns are reordered (for display only) to group each level with its most common partner, so that nested structure appears as a block-diagonal pattern and crossed structure appears as a dense or scattered rectangle.
• upper triangle (lower if swap_triangles=true): a text badge classifying the same pair as one factor nested in the other (A ⊂ B), identical (the two names partition observations the same way), or crossed — complete if every combination of levels is observed, partial (with the observed density) otherwise.

gfs, if given, restricts and orders which grouping factors to show (by name or index); otherwise all distinct grouping factors are shown, in the order they first appear in the model.

The classification only depends on the grouping factors' level assignments (ReMat.refs), not on the original data, matching the model-based approach used by upsetplot!.

The mutating method returns the original object.

nestingtable(m::MixedModel, gfs::Union{Symbol,AbstractString,Integer}...)

Return the co-occurrence table underlying nestingplot, in long format: one row per combination of a level of grouping factor A and a level of grouping factor B, for every pair of distinct grouping factors in m (restricted/ordered by gfs, as in nestingplot). Columns:

• factor_a, level_a, factor_b, level_b: the pair of levels
• count: the number of observations sharing that pair of levels (0 for combinations that never co-occur — these absences are what reveal nesting or partial crossing)

See nestingstructure for the pairwise nested/crossed classification instead of the raw counts.

nestingstructure(m::MixedModel, gfs::Union{Symbol,AbstractString,Integer}...)

Return the pairwise nested/crossed classification underlying nestingplot's upper-triangle badges: one row per pair of distinct grouping factors in m (restricted/ordered by gfs, as in nestingplot), with:

• factor_a, factor_b: the pair of grouping factors
• n_levels_a, n_levels_b: their numbers of levels
• relationship: one of :identical, :a_nested_in_b, :b_nested_in_a, :complete_crossing, or :partial_crossing
• density: the fraction of the n_levels_a × n_levels_b grid of combinations that is actually observed (1.0 for :complete_crossing)

See nestingtable for the raw per-level co-occurrence counts instead of this per-pair summary.

FacetRegressionInfo

Per-group data and within-unit OLS fits underlying facetregression!.

Fields

• group::Symbol: Name of the grouping variable/factor

• predictor::Symbol: Name of the predictor (x) variable

• response::Symbol: Name of the response (y) variable

• labels::Vector: Group levels, in encounter order

• x::Array{Vector{T}, 1} where T<:AbstractFloat: Per-group predictor values

• y::Array{Vector{T}, 1} where T<:AbstractFloat: Per-group response values

• n::Vector{Int64}: Per-group observation counts

• intercept::Vector{T} where T<:AbstractFloat: Per-group within-unit OLS intercept

• slope::Vector{T} where T<:AbstractFloat: Per-group within-unit OLS slope

• xrange::Tuple{T, T} where T<:AbstractFloat: (min, max) of the predictor across all groups

• yrange::Tuple{T, T} where T<:AbstractFloat: (min, max) of the response across all groups

• fixef::Union{Nothing, Tuple{T, T}} where T<:AbstractFloat: Population fixed-effects (intercept, slope); nothing for the table-based constructor

• shrunken_intercept::Union{Nothing, Vector{T}} where T<:AbstractFloat: Per-group shrunken/BLUP intercept; nothing for the table-based constructor

• shrunken_slope::Union{Nothing, Vector{T}} where T<:AbstractFloat: Per-group shrunken/BLUP slope; nothing for the table-based constructor

facetregressioninfo(data, response, predictor, group)::FacetRegressionInfo

Compute the per-group data and within-unit OLS fits underlying facetregression! from a Tables.jl-compatible data source.

response, predictor, and group are Symbol or AbstractString column names. fixef, shrunken_intercept, and shrunken_slope are nothing for this table-based method (there is no model to derive them from).

facetregressioninfo(m::LinearMixedModel, predictor, group=first(fnames(m)))::FacetRegressionInfo
facetregressioninfo(m::LinearMixedModel; group=first(fnames(m)))::FacetRegressionInfo

Compute the per-group data and within-unit OLS fits underlying facetregression! from a fitted LinearMixedModel, sourced entirely from the model's own matrices (m.X, m.y, m.reterms) rather than a raw data table – see ranefinfo (caterpillar.jl) and the co-occurrence extraction in nesting.jl for the precedent of using only ReMat.refs/.levels this way. Also populates fixef (the population intercept/slope) and, per group, shrunken_intercept/shrunken_slope (fixed effects + that group's conditional modes from ranef(m)).

predictor may be omitted, in which case it defaults to the model's only non-intercept fixed-effect term – this throws an ArgumentError if the model has more than one such term.

facetregression(data, response, predictor, group; kwargs...)::Figure
facetregression!(f::Union{GridLayoutBase.GridLayout, GridLayoutBase.GridPosition, GridLayoutBase.GridSubposition, Makie.Figure}, data, response, predictor, group; kwargs...)
facetregression(m::LinearMixedModel, predictor, group=first(fnames(m)); kwargs...)::Figure
facetregression!(f::Union{GridLayoutBase.GridLayout, GridLayoutBase.GridPosition, GridLayoutBase.GridSubposition, Makie.Figure}, m::LinearMixedModel, predictor,
                 group=first(fnames(m)); kwargs...)
facetregression(m::LinearMixedModel; group=first(fnames(m)), kwargs...)::Figure
facetregression!(f::Union{GridLayoutBase.GridLayout, GridLayoutBase.GridPosition, GridLayoutBase.GridSubposition, Makie.Figure}, m::LinearMixedModel;
                 group=first(fnames(m)), kwargs...)
facetregression(info::FacetRegressionInfo; kwargs...)::Figure
facetregression!(f::Union{GridLayoutBase.GridLayout, GridLayoutBase.GridPosition, GridLayoutBase.GridSubposition, Makie.Figure}, info::FacetRegressionInfo; kwargs...)

Create a Cleveland-trellis-style small-multiples display: one panel per level of group, each showing a scatter of response against predictor plus that group's own OLS regression line.

The table-based methods take response, predictor, and group as Symbol or AbstractString column names into any Tables.jl-compatible data.

The LinearMixedModel-based methods instead extract everything from the model's own matrices — no data table is needed. predictor must be the name of a continuous fixed-effect term (e.g. :days); group names a grouping factor and defaults to the first one (first(fnames(m))). Two extra reference lines are drawn in every panel: the population-level fixed-effects fit (show_fixef, the same line in every panel) and that group's own shrunken/BLUP fit (show_shrunken`).

predictor may be omitted entirely (note group then becomes keyword-only, to avoid ambiguity with the predictor positional argument), in which case it defaults to the model's only non-intercept fixed-effect term — this throws an ArgumentError if the model has more than one such term.

Alternatively, facetregressioninfo may be used to construct the FacetRegressionInfo object directly. Constructing it directly can be used to avoid re-computing the per-group OLS fits (and, for a model source, fixef/ ranef) when generating multiple plots (e.g. with different layout or bank45 settings) from the same data.

All panels share linked x/y axes (Cleveland-trellis convention), so slopes and scatter remain visually comparable across panels.

Keywords

• orderby::Union{Symbol,Nothing}=nothing panel order — nothing preserves the order groups are first encountered (not alphabetical), :intercept/:slope sort by the group's own within-unit OLS fit.
• rev::Bool=false: reverse the panel order produced by orderby (applied after sorting, so it also reverses nothing's first-encountered order).
• layout::Union{Nothing,Tuple{Union{Nothing,Int},Union{Nothing,Int}}}=nothing: (nrow, ncol) grid shape. nothing auto-computes a roughly square grid. Either element may be nothing to auto-compute just that one from the other (e.g. layout=(2, nothing) fixes 2 rows and picks enough columns to fit every group). If both are given and their product is smaller than the number of groups, the grid is used as given (never enlarged) and a warning reports how many trailing groups are left undisplayed.
• bank45::Bool=true: shape the grid columns (via colsize! with a GridLayoutBase.Aspect size) so that a line with the mean of the per-group within-unit OLS slopes appears at 45° on screen ("banking", after Cleveland's trellis displays), easing visual comparison of slopes across panels. Full data is always shown — banking reshapes the panels, it never clips data. bank45=false leaves panels at their natural (roughly square) shape.
• xlabel/ylabel: shared axis titles spanning the whole grid (default to the predictor/response names).
• scattercolor, linecolor: per-panel scatter and within-unit regression-line colors.
• show_fixef::Bool=true, fixefcolor=:black, fixeflinestyle=:dash: whether to draw the population fixed-effects line (LinearMixedModel source only).
• show_shrunken::Bool=true, shrunkencolor=:green: whether to draw the per-group shrunken/BLUP line (LinearMixedModel source only).
• labelcolor, labelsize: styling for the per-panel group-level corner label.

The mutating methods return the original object.

facetregressioninfotable(info::FacetRegressionInfo; orderby::Union{Symbol,Nothing}=nothing, rev::Bool=false)
facetregressioninfotable(data, response, predictor, group; orderby::Union{Symbol,Nothing}=nothing, rev::Bool=false)
facetregressioninfotable(m::LinearMixedModel, predictor, group=first(fnames(m));
                         orderby::SUnion{Symbol,Nothing}=nothing, rev::Bool=false)
facetregressioninfotable(m::LinearMixedModel; group=first(fnames(m)),
                         orderby::Union{Symbol,Nothing}=nothing, rev::Bool=false)

Return the per-group OLS fits in info (or derived from data/m) underlying facetregression! as a column table, one row per level of group, with columns:

• group: the group level
• n: number of observations in that group
• intercept, slope: the group's within-unit OLS fit of response on predictor

As with facetregression!, predictor may be omitted for the LinearMixedModel methods (group then becomes keyword-only), defaulting to the model's only non-intercept fixed-effect term.

The LinearMixedModel-derived info additionally has:

• fixef_intercept, fixef_slope: the population fixed-effects fit (same for every row)
• shrunken_intercept, shrunken_slope: that group's shrunken/BLUP fit (fixed effects plus that group's conditional modes from ranef(m))

orderby and rev have the same meaning as in facetregression!.

The data/m-based methods are convenience wrappers equivalent to facetregressioninfotable(facetregressioninfo(data_or_m, args...); orderby, rev).

splom!(f::Indexable, df::DataFrame)

Create a scatter-plot matrix in f from the columns of df.

Non-numeric columns are ignored.

splomaxes!(f::Indexable, labels::AbstractVector{<:AbstractString},
           panel!::Function, args...;
           extraticks::Bool=false, kwargs...)

Populate f with a set of (k*(k-1))/2 axes in a lower triangle for all pairs of labels, where k is the length of labels. The panel! function should have the signature panel!(ax::Axis, i::Integer, j::Integer, args...; kwargs...) and should draw the [i,j] panel in ax.