-------- Original Message -------- Subject: Re: error in digitising landmarks Date: Thu, 3 Feb 2011 13:59:52 -0500 From: Carmelo Fruciano <[email protected]> To: [email protected]
Dear you, I am new to GM, I am working with a hand bone (third metacarpal) and I have chosen 9 landmarks, I also included a ruler on the pictures a scale and gave 2 landmarks more to mark 10mm. I took one picture and copied this file 29 times, digitised the landmarks with TPSdig, made one file with the 30 observations. Then I used MorphoJ and did a procrustes fit then find outliers and the maximum distance was 0.00011. What I really want to do is to calculate de error after digitising 30 times the next image, to see if my landmarks are reliable, and I am not sure how to do it, I tried to run a procrustes ANOVA, but don't understand why I need to do this, and it asks me for classifiers and I am not sure what classifier to include since they are all the same image.
Dear Claudia, first of all I have to say I'm really impressed that you took the time and made the effort of digitizing 30 times the same bone: I would stick to just a few repetitions but, if you can make the same thing with a reasonable number of different specimens that would be great (other than impressive due to the effort that it requires, even if I've been obsessed with error I never took it to this level!). I think that the idea behind many procedures to evaluate error is to relate the variation within specimens (that is, variation associated with digitization) to variation between specimens (that is, probably the real biological variation you're interested in)...the basic idea being that if variation due to digitization is small relative to biological signal you can be pretty confident about the analysis of biological variation... So you would need to carry out this procedure (or a less stressing one with less repetitions) on more specimens to relate the two sources of variation. With MorphoJ I think you would have to put a classifier for the specimen and a classifier for repetition (but it's ages I don't do this in MorphoJ so I'm not sure I remember right)... However, these procedures calculate how reliable is "overall" your digitization. In your particular case, you might have higher error at certain landmarks, lower error at other landmarks (and you might decide that, in your opinion, landmarks with higher digitization-induced variation must be discarded). For this, you probably want to check variation at each landmark. A simple and "exploratory" approach might be to check plots of your digitized landmarks after Procustes superimposition to see if variation at certain landmarks is too large or biased. Ideally, points would be clustered tightly together and form a "circle" (not an ellipse). With your one specimen-30 repetitions dataset you could use TPSRelw, carry out a Procustes superimposition, then click to display the consensus and choose, in the new window, the menu "Options" turning "on" the option "Points". Of course, you're not guaranteed that patterns of error do not vary among specimens, so I would redo it with a few other specimens to check... I'm saying this because you talk about seeing "if my landmarks are reliable" so probably you are more concerned about choice of landmarks other than "overall error"... I hope this helps Carmelo -- Carmelo Fruciano Dipartimento di Biologia University of Catania Tel. +39 095 7306023 Cell. +39 349 5822831 e-mail [email protected] ---------------------------------------------------------------- Universita' di Catania - A.P.Se.Ma. Servizio di Posta Elettronica
