Add columns to _not_ adjust for in `data_adjust()`

[mattansb]

In limma::removeBatchEffect() there is an option to add columns that should be included in the adjusting model, but not adjusted for.

• This requires that all variables are are adjusted for are centered (continuous with center(); factors with contr.sum).
• Then the coefficients for the non-adjusted variables are set to 0 before adjusting.

[etiennebacher]

I don't understand, isn't this what exclude is for?

library(datawizard)

attitude |> 
  data_adjust(exclude = "advance") |> 
  head()
#>        rating complaints privileges    learning     raises   critical advance
#> 1  -8.1102953  5.5583770 -15.848949 -2.75102306  0.5742664  15.605502      45
#> 2   1.6472337  0.0646564  -1.422592 -3.06207012 -1.5567655  -2.315781      47
#> 3   1.0605589 -7.5116953  11.174609  5.59808033  4.8603132   8.061801      48
#> 4  -0.2268416  3.8345277  -4.567441  0.03866933 -7.1185324  13.002574      35
#> 5   6.5462010 -1.2420122  -3.051098  0.87312095 -2.7131349   6.500353      47
#> 6 -10.9418499  5.2030745   2.664156 -1.24552098  4.1370346 -21.678382      34

Do you have an example of what you'd like to have (and that we could also add in tests)?

[mattansb]

No, exclude is the inverse of select - they determine which columns adjusted (Y)
effect (maybe change the name?) determines which columns adjusted for / regressed out (X)

So say the new argument is design, we would have a process that looks like this:

1. Fit a model Y ~ {effect, design}
2. Set all design coefficients to 0, so the (partial) effect of design isn't regressed out
3. Compute $Y-bX$ to obtain Y values adjusted for effect

[strengejacke]

So it's like predict() with a data grid that sets values for all "non-adjusts" to zero?

[mattansb]

That's another way to think about it, yes.

[strengejacke]

Currently, adjust() returns the residuals of the model. Now we would calculate the residuals manually by subtracting expected from predicted, where predicted sets non-adjust to zero? That's how I would implement it.