`fe(m)` seems broken compared to FixefEffectModels

[pdeffebach]

fe(m) doesn't return the augmentdf when I think it is supposed to. Here is an MWE

julia> df = DataFrame(a = rand(1:10, 200), x = rand(200), y = rand(200));

julia> m = reg(df, @formula(y ~ x + fe(a)), save = :fe);

julia> fe(m)[1:5, :]
5×1 DataFrame
 Row │ fe_a     
     │ Float64? 
─────┼──────────
   1 │ 0.47084
   2 │ 0.508088
   3 │ 0.446838
   4 │ 0.508088
   5 │ 0.546581

julia> m_nlreg = nlreg(df, @formula(y ~ x + fe(a)), Poisson(), LogLink(), save = [:fe]);

julia> fe(m_nlreg)
0×0 DataFrame

julia> m_nlreg.augmentdf[1:5, :]
5×1 DataFrame
 Row │ fe_a      
     │ Float64   
─────┼───────────
   1 │ -0.749437
   2 │ -0.679168
   3 │ -0.797734
   4 │ -0.679168
   5 │ -0.609455

[pdeffebach]

The bug is super annoying. My augmentdf has only one column. We index

x.augmentdf[!, 2:size(x.augmentdf, 2)]

and 2:1 is an empty range.

FixedEffectModels.jl, by contrast, has the index of the fixed effect as the first column in their fe DataFrame object.

IMO, best course of action is to add the index to augmentdf

[pdeffebach]

The bug is super annoying. My augmentdf has only one column. We index

x.augmentdf[!, 2:size(x.augmentdf, 2)]

and 2:1 is an empty range.

FixedEffectModels.jl, by contrast, has the index of the fixed effect as the first column in their fe DataFrame object.

IMO, best course of action is to add the index to augmentdf

[jmboehm]

Should be fixed by #60 , let me know if not

[pdeffebach]

Sure, but why not add the fixed effect identifiers? It's always scary using hcat rather than join to attach the fixed effects to the data frame after estimation.