FORUM: latest discussions

2012-02-02 08:08:13 in Food item proximate data analyses

Hi Kari, your approach is very appropiate. Two comment to improve the analysis: 1) zero: usualy the "small" amount should be the 65% of the detection limit or threshold. Nevertheless, if the number of zeros is large then is prefereable to apply the modified alr EM-algorithm (Palarea-Albaladejo and Martín-Fernández, 2008, Computer&Geosciences). 2) MANOVA (and related methods): it is preferable to apply these methods to the ilr-transformed data. The clr-transformation produces a singular (non inverse) covariance matrix. Keep us informed! Cheers! Martin

2012-01-11 10:36:56 in I'm looking for a model

Hi! if you are interested in CODA I see that your trouble will be that your response vector Y could have different dimension in some samples. Our CODA techniques assume that the dimension D (number of parts) is the same for all the samples. Could you consider different models for the different cases of dimensions? Cheers!

2011-12-22 08:18:11 in Application of CoDaPack to partial leach data

Dear, I simply don't understand your troubles. Could you send me an extra explanation to josepantoni.martin@udg.edu ? M.Xmas!

2011-10-19 08:23:32 in 3d graphics

Thank you Martin for your material. I need a bit of time to examine it, also because I'm not at home with MatLab. If I'll experience problems with it, I'll write you again. Thank you very much for your attention. Catia

2011-10-02 13:21:25 in Compositional Analysis of Fruit Fly "Pheromones"

Sorry for the delay, but after holidays I was at meetings and trying to catch up with "things to do". There are no formal procedures to define a SBP (sequential binary partition). I prefer the modelling based approach, because scientists usually know how to interpret the resulting balances inmediately. But if you have absolutely no prior knowledge, you can use the variation matrix (variances of simple logratios) and the biplot to set up something like a modelling approach. Just use your intuition! We are working now on "principal balances" (see the proceedings of CoDaWork'11), which should lead to a SBP that is close to principal components.

2011-05-20 19:52:16 in mean, standard deviation and correlation coefficient

My p o v: 1) you are calculting the mean of a variable (e.g. 15) so as to obtain a "center" of v1 and v2 (e.g. first variable). The point is that the artimethical mean (e.g. 15) is not in the middle (e.g. 10 and 20) when you are assuming that your data is "scale relative" (percentages), i.e. compositional. Note that 15/10 is different than 20/15, thus 15 is not in the "center" between 10 and 20. You should compute the geometrical mean. 2) if you are looking for a "distance" or a "similarity" between compositions you should use an appropriate measure (e.g. Aitchison distance) between samples (compositions). cheers1 Martin

2011-05-01 22:23:24 in IAMG conference proceedings

Thank you again, no worries for the reply. See you there

2011-04-13 15:47:16 in test whether the mean is equal to zero

from my point of view 1) the normality is due to the use of t-test 2) when we work with ilr we can apply all the multivariate statistical methods with no restrictions and the go back with the ilr inverse. This is why ilr can be better.

2011-02-21 08:12:35 in Greetings. About compositional regression.

Hello. Prof, I interest to learn more about the regression of compositional data and testing hypothesus about two groups of compositional data for my thesis but I have difficult to get references about it. Could you give it to me or maybe the link to I get it. Thank you for your help. Kartika

2011-02-17 12:32:20 in (Re)discovering Galton (geometric mean and more)

The paper by Galton is really interesting. Using very simple arguments, he makes a criticism on the use of the normal distribution when measurements are positive.

2011-02-07 12:11:25 in Welcome CODA_colleagues! Your suggestions are needed!

Yes! Very good idea! Send us all your info and we'll include them! Cheers! Martin

Welcome to the CoDa-family! Share with us the progress in CoDa!

Image of CoDaCourses.
                            We are a CoDa-Gang!

Notice board                                   Seminars


July_2012: a week CoDaCourse in Girona!

PhD in Girona: South-East Asian Universities !!

A new CoDa book!!! "Compositional Data Analysis: Theory and Applications"

CoDaWork-2013 will be in Vorau (Austria)!!