Correct. However, the consequence of not rescaling when using Procrustes coordinates is probably not great. For example,
> preds <- shape.predictor(Y.gpa$coords, x = PC, Intercept = TRUE,
+ pred1 = min(PC), pred2 = max(PC)) # PC 1 extremes, more technically
> preds2 <- lapply(preds, geomorph:::cs.scale)
> two.b.pls(preds[[1]], preds2[[1]])
Random PLS calculations: 1000 permutations.
|==========================================================| 100%
Call:
two.b.pls(A1 = preds[[1]], A2 = preds2[[1]])
r-PLS: 1
Effect Size (Z): 3.1783
P-value: 0.001
Based on 1000 random permutations
> sapply(preds, geomorph:::csize)
pred1 pred2
1.004162 1.001530
> sapply(preds2, geomorph:::csize)
pred1 pred2
1 1
So by not using cs.scale, the resulting centroid size is nearly unit size and the two approaches — scaling or not scaling — produce perfectly correlated shapes, meaning an affine transformation of one to the other (and I am sure the affine transformation is just enlargement). Although I have not verified, I believe using all PCs instead of just the first would result in predictions that have unit size. The distortion caused by not rescaling is probably far less than would be made if someone made a figure in powerpoint and rescaled the deformation grids that are slapped onto a PC plot. This is to say, this might be much ado about nothing, unless one wishes to predict shapes way beyond the realm of possibilities encompassed by the observed data, which one could do, for example, with picknplot.shape. Not rescaling configurations would be weird, as some deformation grids would expand and others shrink, but it would just be a photocopy enlargement effect.
With boas coordinates, it’s not quite the same, and the photocopy enlargement effect would be important.