Question: Is wTO + addaptive alpha still the best way to handle local dependency between items?

[E-Mendez]

Hi!

I found out I cannot apply the routine proposed in Christensen et al (2020), which is UVA with wTO + Adaptive alpha (1-). Said change was applied in version 1.2.0, to remove dependencies (2-).

( Error in redundancy.process(data = data, cormat = cormat, n = n, model = model, :
"alpha" and "adapt" have been removed from UVA. Please use "threshold" )

My questions are:

1. Is the routine proposed in (1-) still the preferred method? Or I'm behind in the literature?
2. And... is the removal of those options only a computational decision regarding dependency hells aimed at a cleaner code, or again it is based also on theoretical decisions preferring the mode "threshold"?

Thank you for your work!!

Sources

1- Christensen, A. P., Garrido, L. E., & Golino, H. (2020). Unique variable analysis: A novel approach for detecting redundant variables in multivariate data. PsyArXiv, 10.

2- REMOVE: "alpha" and "adapt" options in UVA (removes {fitdistrplus} dependency)

[AlexChristensen]

Hi @E-Mendez,

Based on simulation work, adaptive alpha is not the best way to handle local dependencies.

The main manuscript for UVA was just published at Multivariate Behavioral Research: https://www.tandfonline.com/doi/abs/10.1080/00273171.2023.2194606

Attaching preprint for direct access: Unique Variable Analysis_preprint.pdf

Further: The defaults are to use a cut-off with cut.off = 0.25 and reduce.method = "remove"

[AlexChristensen]

See commit referencing this issue

Support for auto = FALSE, type = "alpha" (adaptive alpha), and type = "alpha" will no longer be provided (bugs will not be fixed and warnings will be thrown)

The goal is to move toward best practices that are published in MBR (see citation in thread)

Closing with commit