Hi @gowerc,
Currently, rbmi:::ancova_single() restricts the treatment factor to two levels only and therefore requires pairwise subsetting in multi-arm trials data.
In studies with more than two treatment arms, it might be preferable to fit the ANCOVA model using all available subjects and then extract the relevant pairwise LS-mean contrasts. This would allow information from all treatment arms to contribute to the estimation of nuisance parameters (e.g., covariate effects and residual variance), rather than discarding observations from the non-compared arms.
Would it be worth considering support for multi-level treatment factors and post-hoc contrasts, rather than requiring a binary treatment variable? Or is there a specific reason for the current implementation that I may be overlooking?
Thanks for any comments.
PS. This might be related to #520
Hi @gowerc,
Currently,
rbmi:::ancova_single()restricts the treatment factor to two levels only and therefore requires pairwise subsetting in multi-arm trials data.In studies with more than two treatment arms, it might be preferable to fit the ANCOVA model using all available subjects and then extract the relevant pairwise LS-mean contrasts. This would allow information from all treatment arms to contribute to the estimation of nuisance parameters (e.g., covariate effects and residual variance), rather than discarding observations from the non-compared arms.
Would it be worth considering support for multi-level treatment factors and post-hoc contrasts, rather than requiring a binary treatment variable? Or is there a specific reason for the current implementation that I may be overlooking?
Thanks for any comments.
PS. This might be related to #520