PCA versus ANCOVA

[marlneil]

One of my college mates is facing a problem. Their adviser is urging them to use a PCA for their data set (the adviser uses PCA for everything) but the adviser won't explain how to use PCA or how to interpret the results. My college mate's data set is very simple and is just testing whether groups of muscles are larger significantly between two types of organisms. I think something simple like running an ANCOVA for each of the groups being compared would make more sense and give easy to interpret results, which would greatly benefit my college mate, since they don't have a heavy stats background and since their project basically revolves around the question of, "Are the muscle group larger in organism A versus organism B?"

Something this simple makes more sense to me to analyze this way- http://udel.edu/~mcdonald/statancova.html like with the example given of Prieto-Marquez et al.'s data set, using some body dimension as a covariate for different muscle groups. I know several related papers which have tackled this the same way in the literature.

So what I'm wondering is why choose a PCA here when ANCOVA seems like it would be simpler all around? Is my adviser using PCA correctly or just abusing it?

Thank you.

[lumhearts]

There is probably some confusion somewhere regarding the objectives of your "friend's" study. PCA and ANOVA address two completely different types of research questions. They also require different types of data.

So to answer your question - I don't think there is any reason to select PCA to analyze the stated research question. Giving the advisor the benefit of the doubt (which they typically have earned), he/she may be trying to suggest a more complex and interesting research project.

[bugman]

There may be a case for considering MANOVA if you have "groups" of muscles or multiple muscle measurements per case, since there is probably some degree of dependance among your (or your mates) dependant varaibles.

PCA might be useful as a exploratory tool if you have several muscle types or groups per individual, but for the formal/hypothesis testing component of this research, then as lumhearts said, ANOVA (or MANOVA depending on the nayure of your data) is most appropriate.