Taking weighting seriously

src/lm.jl

@@ -87,10 +87,10 @@
     return -n / 2 * (log(2π * deviance(r) / n) + 1)
 end
 
 function loglikelihood(r::LmResp{T,<:AnalyticWeights}) where {T}
     N = length(r.y)
     n = sum(log, weights(r))
     return (n - N * (log(2π * deviance(r) / N) + 1)) / 2
 end
 
 function loglikelihood(r::LmResp{T,<:ProbabilityWeights}) where {T}
@@ -169,7 +169,7 @@
                      :fit)
         fweights(wts)
     else
         throw(ArgumentError("`wts` should be an `AbstractVector` coercible to `AbstractWeights`"))
     end
 end
 
@@ -178,12 +178,12 @@
              dropcollinear::Bool=true, method::Symbol=:qr)
     # For backward compatibility accept wts as AbstractArray and coerce them to FrequencyWeights
     _wts = convert_weights(wts)
-    if !(wts isa AbstractWeights && isempty(_wts))
+    if isempty(_wts)
         Base.depwarn("Using `wts` of zero length for unweighted regression is deprecated in favor of " *
                      "explicitly using `UnitWeights(length(y))`." *
                      " Proceeding by coercing `wts` to UnitWeights of size $(length(y)).",
                      :fit)
         _wts = uweights(length(y))
     end
 
     if method === :cholesky
@@ -202,7 +202,7 @@
              contrasts::AbstractDict{Symbol}=Dict{Symbol,Any}())
     f, (y, X) = modelframe(f, data, contrasts, LinearModel)
     _wts = convert_weights(wts)
-    _wts = !(wts isa AbstractWeights) && isempty(_wts) ? uweights(length(y)) : _wts
+    _wts = isempty(_wts) ? uweights(length(y)) : _wts
     if method === :cholesky
         fit!(LinearModel(LmResp(y, _wts), cholpred(X, dropcollinear, _wts), f))
     elseif method === :qr