missing type leads to errors

[felixholub]

On julia-1.2.0, FixedEffectModels.jl version [9d5cd8c9] FixedEffectModels v0.8.2 I get the following error messages when including a variable which allows for missings. Take the following example:

using DataFrames, FixedEffectModels

df1 = DataFrame(a=[1.0, 2.0, 3.0, 4.0], b=[5.0, 7.0, 11.0, 13.0])
df2 = DataFrame(a=[1.0, missing, 3.0, 4.0], b=[5.0, 7.0, 11.0, 13.0])

rr1a = reg(df1,@model(a ~ b))
rr1b = reg(df1,@model(b ~ a))
rr2a = reg(df2,@model(a ~ b))
rr2b = reg(df2,@model(b ~ a))

The first two regressions work well, while the third produces

julia> rr2a = reg(df2,@model(a ~ b))

ERROR: MethodError: no method matching Array{Float64,1}(::Array{Float64,2})
Closest candidates are:
  Array{Float64,1}(::AbstractArray{S,N}) where {T, N, S} at array.jl:482
  Array{Float64,1}() where T at boot.jl:423
  Array{Float64,1}(::UndefInitializer, ::Int64) where T at boot.jl:404
  ...
rr2a = reg(df2,@model(a ~ b))

ERROR: MethodError: no method matching Array{Float64,1}(::Array{Float64,2})
Closest candidates are:
  Array{Float64,1}(::AbstractArray{S,N}) where {T, N, S} at array.jl:482
  Array{Float64,1}() where T at boot.jl:423
  Array{Float64,1}(::UndefInitializer, ::Int64) where T at boot.jl:404
  ...
Stacktrace:
 [1] convert(::Type{Array{Float64,1}}, ::Array{Float64,2}) at ./array.jl:474
 [2] #reg#38(::Nothing, ::Expr, ::Nothing, ::Nothing, ::Int64, ::Dict{Symbol,Any}, ::Float64, ::Int64, ::Bool, ::Symbol, ::Bool, ::typeof(reg), ::DataFrame, ::StatsModels.FormulaTerm{StatsModels.Term,StatsModels.Term}) at /home/holub/.julia/packages/FixedEffectModels/kANnE/src/reg.jl:158
 [3] reg(::DataFrame, ::StatsModels.FormulaTerm{StatsModels.Term,StatsModels.Term}) at /home/holub/.julia/packages/FixedEffectModels/kANnE/src/reg.jl:49
 [4] #reg#37(::Base.Iterators.Pairs{Union{},Union{},Tuple{},NamedTuple{,Tuple{}}}, ::typeof(reg), ::DataFrame, ::Model) at /home/holub/.julia/packages/FixedEffectModels/kANnE/src/reg.jl:36
 [5] reg(::DataFrame, ::Model) at /home/holub/.julia/packages/FixedEffectModels/kANnE/src/reg.jl:36
 [6] top-level scope at none:0

and the fourth produces

rr2b = reg(df2,@model(b ~ a))

ERROR: ArgumentError: FDist: the condition ν1 > zero(ν1) && ν2 > zero(ν2) is not satisfied.
Stacktrace:
 [1] macro expansion at /home/holub/.julia/packages/Distributions/tfkz4/src/utils.jl:6 [inlined]
 [2] Type at /home/holub/.julia/packages/Distributions/tfkz4/src/univariate/continuous/fdist.jl:30 [inlined]
 [3] Type at /home/holub/.julia/packages/Distributions/tfkz4/src/univariate/continuous/fdist.jl:35 [inlined]
 [4] Type at /home/holub/.julia/packages/Distributions/tfkz4/src/univariate/continuous/fdist.jl:36 [inlined]
 [5] compute_Fstat(::Array{Float64,1}, ::LinearAlgebra.Symmetric{Float64,Array{Float64,2}}, ::Int64, ::Bool, ::VcovSimpleMethod, ::VcovData{LinearAlgebra.Cholesky{Float64,Array{Float64,2}},1}) at /home/holub/.julia/packages/FixedEffectModels/kANnE/src/vcov/utils.jl:14
 [6] #reg#38(::Nothing, ::Expr, ::Nothing, ::Nothing, ::Int64, ::Dict{Symbol,Any}, ::Float64, ::Int64, ::Bool, ::Symbol, ::Bool, ::typeof(reg), ::DataFrame, ::StatsModels.FormulaTerm{StatsModels.Term,StatsModels.Term}) at /home/holub/.julia/packages/FixedEffectModels/kANnE/src/reg.jl:363
 [7] reg(::DataFrame, ::StatsModels.FormulaTerm{StatsModels.Term,StatsModels.Term}) at /home/holub/.julia/packages/FixedEffectModels/kANnE/src/reg.jl:49
 [8] #reg#37(::Base.Iterators.Pairs{Union{},Union{},Tuple{},NamedTuple{,Tuple{}}}, ::typeof(reg), ::DataFrame, ::Model) at /home/holub/.julia/packages/FixedEffectModels/kANnE/src/reg.jl:36
 [9] reg(::DataFrame, ::Model) at /home/holub/.julia/packages/FixedEffectModels/kANnE/src/reg.jl:36
 [10] top-level scope at none:0

Previously, I experienced that missing where simply dropped for the regression.

[matthieugomez]

Thanks for the MWE. This is due to JuliaStats/StatsModels.jl#145.
I have pushed a fix, and tagged a new version.