Re: [R-sig-eco] probability distribution for zero-inflated, right skewed data
Definitely looks promising. Thanks Bob! On Jun 16, 2014, at 2:29 PM, Bob O'Hara wrote: On 16/06/14 13:57, Johannes Björk wrote: Dear all, Im looking into how to fit a GLM model (Im using rjags) with data that are heavily right skewed. In addition, some variables also zero-inflated. The data are species area distribution measured as total area (km^2) which is subsetted into area in tropical zone and area in temperate zone. The last two variables contain zeros. I have google zero-inflated models... and most that come up is zero-inflated negative binomial and zero-inflated negative poisson for count data. I reckon I cannot use any of these distributions since my variables are not discrete. Any pointer to which distribution(s) that might fit this kind of data would be much appreciated. I think a Tweedie distribution is sometimes used, but that always makes me think of escaping chickens. Recently this was published, which might also be useful: http://onlinelibrary.wiley.com/doi/10./2041-210X.12122/abstract. They used BUGS, so you could ask them if the code is available. Even if it isn't, it shouldn't be too difficult to code up. Bob -- Bob O'Hara Biodiversity and Climate Research Centre Senckenberganlage 25 D-60325 Frankfurt am Main, Germany Tel: +49 69 7542 1863 Mobile: +49 1515 888 5440 WWW: http://www.bik-f.de/root/index.php?page_id=219 Blog: http://blogs.nature.com/boboh Journal of Negative Results - EEB: www.jnr-eeb.org ___ R-sig-ecology mailing list R-sig-ecology@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-ecology ___ R-sig-ecology mailing list R-sig-ecology@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-ecology
Re: [R-sig-eco] probability distribution for zero-inflated, right skewed data
You could try estimating the conditional cumulative distribution function with quantile regression by estimating a large interval of quantiles (e.g., 0.01 to 0.99 if your n is large enough). Quantile regression will readily handle skewed and heterogeneous responses. Some finessing required to check when estimates are above a mass of zeros but this is all doable. Brian Brian S. Cade, PhD U. S. Geological Survey Fort Collins Science Center 2150 Centre Ave., Bldg. C Fort Collins, CO 80526-8818 email: ca...@usgs.gov brian_c...@usgs.gov tel: 970 226-9326 On Mon, Jun 16, 2014 at 5:57 AM, Johannes Björk bjork.johan...@gmail.com wrote: Dear all, Im looking into how to fit a GLM model (Im using rjags) with data that are heavily right skewed. In addition, some variables also zero-inflated. The data are species area distribution measured as total area (km^2) which is subsetted into area in tropical zone and area in temperate zone. The last two variables contain zeros. I have google zero-inflated models... and most that come up is zero-inflated negative binomial and zero-inflated negative poisson for count data. I reckon I cannot use any of these distributions since my variables are not discrete. Any pointer to which distribution(s) that might fit this kind of data would be much appreciated. Attached: Histograms of the data ___ R-sig-ecology mailing list R-sig-ecology@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-ecology [[alternative HTML version deleted]] ___ R-sig-ecology mailing list R-sig-ecology@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-ecology
Re: [R-sig-eco] probability distribution for zero-inflated, right skewed data
Hi, Just a little note about computational methods for Tweedies and compound Poisson-gammas. The package statMod has a Tweedie family that can be used in glm(), the package Tweedie has functions for density calculations (and other things), the package fishMod (my own package) has methods to fit Tweedie GLMs and compound Poisson-gamma, and lastly (but very far from leastly) mgcv has a Tweedie family too. The mgcv fact seems to not be widely known. None of these methods are Bayesian however. As Bob suggested coding up a sampler should be easy enough, either from the compound representation or directly from the (marginal) Tweedie -- see ?rTweedie in the fishMod package or ?rtweedie in the tweedie package. At the risk of blatant self-promotion have a look at my paper on compound Poisson-gammas -- there is a review of some of the more common methods. http://link.springer.com/article/10.1007/s10651-012-0233-0 Good luck! Scott On 16/06/14 22:29, Bob O'Hara wrote: On 16/06/14 13:57, Johannes Björk wrote: Dear all, Im looking into how to fit a GLM model (Im using rjags) with data that are heavily right skewed. In addition, some variables also zero-inflated. The data are species area distribution measured as total area (km^2) which is subsetted into area in tropical zone and area in temperate zone. The last two variables contain zeros. I have google zero-inflated models... and most that come up is zero-inflated negative binomial and zero-inflated negative poisson for count data. I reckon I cannot use any of these distributions since my variables are not discrete. Any pointer to which distribution(s) that might fit this kind of data would be much appreciated. I think a Tweedie distribution is sometimes used, but that always makes me think of escaping chickens. Recently this was published, which might also be useful: http://onlinelibrary.wiley.com/doi/10./2041-210X.12122/abstract. They used BUGS, so you could ask them if the code is available. Even if it isn't, it shouldn't be too difficult to code up. Bob -- Scott Foster Computational Informatics CSIRO E scott.fos...@csiro.au T +61 3 6232 5178 Postal address: CSIRO Computational Informatics, GPO Box 1538, Hobart TAS 7001 Street Address: CSIRO Computational Informatics, Castray Esplanade, Hobart Tas 7001, Australia www.csiro.au ___ R-sig-ecology mailing list R-sig-ecology@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-ecology