Sir Jasper,

You still need to set a scale for capturing images of your specimen. This is to ensure that all images are of the same size in order for the analysis to really remove the element of size in the process.

GM analysis like any traditional morphometric technique still computes the size. However, what makes GM analysis very unique is that you can compute the allometry immediately. Allometry allows the researchers to determine how size can contribute to SHAPE analysis. You may have the option to whether proceed in the analysis analysis with that kind of allometry, but of course this could be deal-breaker in some journals (reviewers) that have a certain threshold for allometry.

The GM process conducts SHAPE analysis but we need to take extra precautions in capturing images in order to remove the effect of size during the analysis.

I hope this was clear. Let me know if you have any more questions. LOL

The reason for setting the scale is to obtain information about size as well as shape. If you do not set the scale factor, centroid size will be in units of pixels

There might be additional reasons: although the aim ofshape analysis is indeed to remove size as a nuisance factor, you might need to rule out whether shape is dependent on it. So you might need to rule out allometry (dependence of shape on size), or to determine whether any allometric signal you find is static, sexual or ontogenetic for which you do indeed need the centroid sizes.

It can be confusing to call "size" a nuisance factor because, in shape analysis, the nuisance factor is "scale". If shape depends on size, that is information, not noise.

I see what you mean. Nevertheless, relative size is standardized as part of the generalized Procrustes procedure, which basically amounts to treating it as noise. In that respect, he seems to have a point, in that from a purely theoretical perspective, scaling is indeed optional (it doesn't matter if scale is removed in pixes or mm). However, since pixel size isn't a particularly informative variable, nor a standardized one (my pixels aren't your pixels), scaling does come warmly recommended. Does that sum it up about right?

My point really is simple. You do not need to set the scale factor in tpsDig to superimpose the data (they will be scaled even if you do not set the scale factor). I am not saying that scaling is optional, what is optional is setting the scale factor. Scaling is optional in another sense, although not one I intended--you have the option to work in size-and-shape space, if you wish. However, in context of the original question, the reason for setting the scale factor is to get centroid size in units other than pixels. The reason for my comment on terminology, above, is that the word "size" has more than one meaning in the morphometrics literature. Because of that ambiguity, it can be confusing to refer to "size" as a nuisance parameter. Similarly, it can be confusing to refer to scaling as "size-standardization" because size-standardization often refers to removing the variation in shape that is due to size. Allometric variation in shape remains in the data following a Procrustes superimposition because it is not a nuisance parameter. It is signal not noise.

To answer this question I think its best way to separate shape and size measurement.

First, size by it self can be measured either by centroid size (the average of length from each point toward the center) or traditional measurement (length from one end to another end).

Second, shape variables can be obtain by Generalized Procrustes analysis (GPA), this mean every figure undergo translation, rotation, and SCALING to obtain shape variables (Procrustes alignment makes all figure have same size, which also mean "removal of size".

The geometric morphometric that I have experience with is CLIC so I am not sure if this is apply to other geometric morphometric program.

To answer the question, do you need to set the scale to perform geometric morphometric analysis, I say "yes". Because changes in size can influence shape. Plotting landmarks or digitization alone will not measure size for you unless you set the scale and you will not be able to obtain shape variables since GPA required it for scaling. My idea is similar to Thaddeus Carvajal is that to ensure that your shape analysis is valid and not being influence by size, "allometry" test should be performed. Allometry make sure that your shape analysis is independent from size variables.