Actually the rule is that the number of specimens should be larger than the number of variables 2p-4 not p landmarks in the case of 2D data.
__________________F. James Rohlf, Distinguished Prof. Emeritus Dept. Anthropology and Ecology & Evolution Stonybrook University -------- Original message --------From: Carmelo Fruciano <c.fruci...@unict.it> Date: 3/12/18 9:51 PM (GMT-10:00) To: MORPHMET <firstname.lastname@example.org> Subject: Re: ***SPAM*** [MORPHMET] Re: Digitizing landmarks on live larvae Dear Avi, I guess it's hard to formulate any rule of thumb. Some analyses might not be defined if the number of variables exceed the number of observations. Some other analyses might be defined (there is many distance-based analyses nowadays which circumvent that problem). However, whether a given analysis is defined doesn't mean that the inference is necessarily accurate. I guess it depends on what you plan to test/measure and the effect size you expect. Andrea Cardini has some papers that can give you some hint on this, such as: Cardini & Elton 2007 - Zoomorphology Cardini et al. 2015 - Zoomorphology Cardini & Elton 2017 - Hystrix They should be downloadable from his website https://sites.google.com/site/alcardini/home/pubs Obviously, having an idea of variation on your own data would be better. I hope this helps. Carmelo On 3/12/2018 8:27 PM, Avi Koplovich wrote: Hi Carmelo, Thank you for those answers. One more question please: I know that the number of specimens should exceed the number of the total landmarks (fixed-landmarks + semi-landmarks). Is there a rule of thumb of by how much or what ratio between specimens to semi-landmarks one should keep? Thank you, Avi On Sunday, March 11, 2018 at 5:15:24 PM UTC+2, Carmelo Fruciano wrote: Il 6/03/2018 4:44 PM, Avi Koplovich ha scritto: > Hi Carmelo, > Thank you for your answer. > My project tests for the influence of kairomones of a predator fish on > the morphology of Salamander larvae during its development. To do this, > I take pictures every other week of larvae spawned from six different > females and assigned to 3 treatments: No fish, 3 caged fish, 6 caged fish. Hi Avi, it sounds like an interesting experiment. I will try to answer to your questions but keeping in mind that I'm not very knowledgeable on salamander development. > 1. I intend to use landmark 1 (dorsal connection of the tail fin) as a > fixed factor. But I thought I may be able to use the tail tip > (landmark 20) and head tip (landmark 48) as fixed landmarks as well. > Do you think it's ok in an ontogeny experiment? I guess it will depend on how long into ontogeny you will track the larvae and whether or not that point will "disappear" over ontogeny and/or slide unreasonably (depends also on your question). You, being knowledgeable on their biology, are the best judge on that. > If not, do you think > it's ok to slide all semi-landmarks of the tail on landmark 1, and > all head semi-landmarks on an eye landmark? Since the eye isn't part > of the head contour, is it ok if I slide one semi-landmark to the > eye and all rest semi-landmarks of the head one to each other as a > closed shape? The point(s) slid relative to the eye won't be sliding along the direction tangent to the curve you want to approximate (i.e., the curvature of the head). A good starting point on the method could be Gunz & Mitteroecker 2013 - Hystrix > 2. Is it ok if landmarks 1 and 39 slid relative to each other as well > as 41 and 55, since both describe a closed shape? It's not particularly desirable (see answer above). > 3. Another worry I have is that landmark 40 which I used to create the > comb fan for both the tail and the head is too far from both of them > so it doesn't bypass the bending. > 4. I'm affraid I don't fully understand why landmark 40 can not be > treated as a fixed landmark. In the book of Zelditch 2004, she says > that one of the basic differences between fixed-landmark and > semi-landmark is the degree of freedom, while fixed has two because > it is docked on both X and Y axes while semi only on one of them > (depending on the nature of the specific fan). Please correct me if > I'm wrong, but what if I use the side line of the larvae (which is > an anatomical/homologous feature) as my X axis and use the Y > component of landmark 1 (dorsal connection of the tail fin) to dock > landmark 40 on the Y axis? Is it wrong because of the dependency of > landmark 40 on landmark 1 regarding the Y coordinate? I think Don has covered these two very well. > 5. Emma Sherratt told me she straightened the bent tail-body using TPS > software in her paper Sherratt et al. 2017 - Nature ecology & > evolution. In the supplementary material of her paper she wrote: > "To correct for dorso-ventral bending in the landmark configurations > (caused by the joint of the tail with the head/body), we used the > ‘unbend specimens’ function of tpsUtil v.1.86 (Rohlf 2015). The > landmark configurations for each specimen were transformed using the > quadratic approach, straightening from the eye (1) along the > notochord landmarks (46 to 55) to the tip of the tail (8)." > Jim mentioned this unbending function here before. I read the help > about unbending specimens and thought I can use landmarks 20 (tail > tip), 48 (head tip) and several semi-landmarks I can digitize using > the comb fan (equally spaced) along the side line of the larvae, in > order to create the quadratic curve (while the side line "helper" > semi-landmarks can be later omitted from the dataset - I saw > Fruciano et al. 2016). Does this sound good? > I bet that this can basically solve the problems I mentioned in 3 & > 4, since then I can digitized the whole body contour. That function is great but, as everything, relies on a set of assumptions (see also Fruciano 2016 - Development Genes and Evolution for a brief discussion). In your case: - that you can consistently identify those points along an hypothetical line (which you would remove after the unbending) - that your arching is well represented by the chosen function You are the best judge on whether these assumptions are satisfied in your case or not. I suggested the Valentin et al. (2008 - Journal of Fish Biology) approach because it's more flexible (less stringent assumptions, which obviously doesn't mean assumption-free) and therefore more generally applicable. But, of course, the approach in tpsUtil can be a great solution if it's appropriate to your data. Best, Carmelo -- MORPHMET may be accessed via its webpage at http://www.morphometrics.org --- You received this message because you are subscribed to the Google Groups "MORPHMET" group. To unsubscribe from this group and stop receiving emails from it, send an email to morphmet+unsubscr...@morphometrics.org.