`get_data()` does not return scaled variables in subset.

[strengejacke]

Works with log() etc., but not with scale(). Haven't looked into it in details, but looks a bit more complicated.

library(insight)
m <- lm(log(Sepal.Length) ~ scale(Sepal.Width), data = iris, subset = Species == "versicolor")
out <- get_data(m)

head(out)
#>    Sepal.Length Sepal.Width    Species
#> 51          7.0  0.32731751 versicolor
#> 52          6.4  0.32731751 versicolor
#> 53          6.9  0.09788935 versicolor
#> 54          5.5 -1.73753594 versicolor
#> 55          6.5 -0.59039513 versicolor
#> 56          5.7 -0.59039513 versicolor

head(subset(iris, 
            subset = Species == "versicolor", 
            select = c("Sepal.Length", "Sepal.Width", "Species")))
#>    Sepal.Length Sepal.Width    Species
#> 51          7.0         3.2 versicolor
#> 52          6.4         3.2 versicolor
#> 53          6.9         3.1 versicolor
#> 54          5.5         2.3 versicolor
#> 55          6.5         2.8 versicolor
#> 56          5.7         2.8 versicolor




m <- lm(log(Sepal.Length) ~ log(Sepal.Width), data = iris, subset = Species == "versicolor")
out <- get_data(m)

head(out)
#>    Sepal.Length Sepal.Width    Species
#> 51          7.0         3.2 versicolor
#> 52          6.4         3.2 versicolor
#> 53          6.9         3.1 versicolor
#> 54          5.5         2.3 versicolor
#> 55          6.5         2.8 versicolor
#> 56          5.7         2.8 versicolor

head(subset(iris, 
            subset = Species == "versicolor", 
            select = c("Sepal.Length", "Sepal.Width", "Species")))
#>    Sepal.Length Sepal.Width    Species
#> 51          7.0         3.2 versicolor
#> 52          6.4         3.2 versicolor
#> 53          6.9         3.1 versicolor
#> 54          5.5         2.3 versicolor
#> 55          6.5         2.8 versicolor
#> 56          5.7         2.8 versicolor

^{Created on 2022-03-21 by the reprex package (v2.0.1)}

[bwiernik]

Is the issue because scale() returns a matrix instead of a vector?

[strengejacke]

I think so. It's treated as "matrix column", and not processed by .backtransform(). However, since we have the center and scale as attributes, maybe we should just backtransform "scale" as well?

[bwiernik]

Yes I think so. We should convert the matrix column back to a standard vector column too

[strengejacke]

We should convert the matrix column back to a standard vector column too

That's already done.

[strengejacke]

Using subset drops the scale- and center attributes, that's why it doesn't work:

m <- lm(log(Sepal.Length) ~ scale(Sepal.Width), data = iris, subset = Species == "versicolor")
attributes(model.frame(m)$`scale(Sepal.Width)`)
#> $dim
#> [1] 50  1

m <- lm(log(Sepal.Length) ~ scale(Sepal.Width), data = iris)
attributes(model.frame(m)$`scale(Sepal.Width)`)
#> $dim
#> [1] 150   1
#> 
#> $`scaled:center`
#> [1] 3.057333
#> 
#> $`scaled:scale`
#> [1] 0.4358663

That's indeed tricky... Maybe we should - for now - at least give a warning?

[vincentarelbundock]

Should we close this in favor of #691 ?

[strengejacke]

Let me check the examples in this issue and if they work with #691. If so, we can close