In his text, Wheeler advocates for using the modulus F-statistic when exploring influences on the loss function because...

1- the modulus F-statistic is associated with higher statistical power

2- when using the modulus F-statistic, the probability of committing a Type I error is reduced

3- Fisher's F-statistic assumes equal variances but that assumption doesn't hold (or is, at least, unproven)

4- Fisher's F-statistic can't be used in the absence of replication