Residuals in GLM

.github/workflows/CI-future.yml

@@ -24,7 +24,7 @@ jobs:
         arch: ['x64']
     steps:
       - run: git config --global core.autocrlf false
-      - uses: actions/checkout@v3
+      - uses: actions/checkout@v4
       - uses: julia-actions/setup-julia@v1
         with:
           version: ${{ matrix.version }}

.github/workflows/CI-stable.yml

@@ -24,7 +24,7 @@ jobs:
         arch: ['x64']
     steps:
       - run: git config --global core.autocrlf false
-      - uses: actions/checkout@v3
+      - uses: actions/checkout@v4
       - uses: julia-actions/setup-julia@v1
         with:
           version: ${{ matrix.version }}

.github/workflows/Documenter.yml

@@ -13,7 +13,7 @@ jobs:
     name: Documentation
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v3
+      - uses: actions/checkout@v4
       - uses: julia-actions/julia-buildpkg@latest
       - uses: julia-actions/julia-docdeploy@latest
         env:

Project.toml

@@ -27,7 +27,7 @@ SpecialFunctions = "0.6, 0.7, 0.8, 0.9, 0.10, 1, 2.0"
 StatsAPI = "1.4"
 StatsBase = "0.33.5, 0.34"
 StatsFuns = "0.6, 0.7, 0.8, 0.9, 1.0"
-StatsModels = "0.6.23, 0.7"
+StatsModels = "0.7.3"
 Tables = "1"
 julia = "1.6"
 

src/GLM.jl

@@ -10,6 +10,7 @@ module GLM
     import Base: (\), convert, show, size
     import LinearAlgebra: cholesky, cholesky!
     import Statistics: cor
+    using StatsAPI
     import StatsBase: coef, coeftable, coefnames, confint, deviance, nulldeviance, dof, dof_residual,
                       loglikelihood, nullloglikelihood, nobs, stderror, vcov,
                       residuals, predict, predict!,
@@ -21,7 +22,7 @@ module GLM
     export coef, coeftable, confint, deviance, nulldeviance, dof, dof_residual,
            loglikelihood, nullloglikelihood, nobs, stderror, vcov, residuals, predict,
            fitted, fit, fit!, model_response, response, modelmatrix, r2, r², adjr2, adjr²,
-           cooksdistance, hasintercept, dispersion
+           cooksdistance, hasintercept, dispersion, vif, gvif, termnames
 
     export
         # types

src/glmfit.jl

@@ -844,6 +844,7 @@ end
 const _RESIDUAL_TYPES = [:deviance, :pearson, :response, :working]
 
 """
+    residuals(model::LinearModel; type=:deviance)
     residuals(model::GeneralizedLinearModel; type=:deviance)
 
 Return the residuals of a GLM.
@@ -852,7 +853,7 @@ Supported values for `type` are:
 - `:deviance` (the default): the signed square root of the element-wise
   contribution to the deviance
 - `:response`: the difference between the observed and fitted values
-- `:working`: working residuals (used during the IRLS process)
+- `:working`: working residuals (used during the IRLS process). For a linear model, it is same as `:response`. 
 - `:pearson`: Pearson residuals, i.e., response residuals scaled
     by the standard error of the response
 """

src/linpred.jl

@@ -360,9 +360,21 @@ response(obj::LinPredModel) = obj.rr.y
 
 fitted(m::LinPredModel) = m.rr.mu
 predict(mm::LinPredModel) = fitted(mm)
-residuals(obj::LinPredModel) = residuals(obj.rr)
 
-function formula(obj::LinPredModel)
+function residuals(model::LinPredModel; type=:deviance)
+    type in _RESIDUAL_TYPES ||
+        throw(ArgumentError("Unsupported type `$(type)``; supported types are" *
+                            "$(_RESIDUAL_TYPES)"))
+
+    resid = response(model) - fitted(model)
+    if length(model.rr.wts) > 0 && (type === :deviance || type === :pearson)
+        return resid .* sqrt.(model.rr.wts)
+    else
+        return resid
+    end
+end
+
+function StatsModels.formula(obj::LinPredModel)
     obj.formula === nothing && throw(ArgumentError("model was fitted without a formula"))
     return obj.formula
 end

src/lm.jl

@@ -70,8 +70,6 @@ function loglikelihood(r::LmResp)
     -n/2 * (log(2π * deviance(r)/n) + 1)
 end
 
-residuals(r::LmResp) = r.y - r.mu
-
 """
     LinearModel