Re: [R-sig-eco] forward selection RDA after controlling for constraints
Dear Stephen, I assume that your approach will account for spatial structure (large scale spatial trend), i.e. remove spatial structure prior to analysis and hence also remove spatially structured but ecologically potentially important variables. This approach, however, does not necessarily remove spatial autocorrelation (which can be thought of some sort of distance decay). Don't mind, it often gets confused. I do not know any straightforward way, unfortunately, to control for spatial autocorrelation in multivariate analyses (unless some autoregressive models or GEE would be implemented in RDA/CCA instead of glm) but I would think that the most promising workaround would be some sort of spatial filtering (see e.g., Diniz-Filho et al. 2003, Dray et al. 2006) as e.g. implemented in function ME {spdep} by Pedro Peres-Neto. Bini et al. (2009) showed that spatial filtering on the residuals is close to what is produced by ME (though Jari may warn against a regression/RDA on residuals). Filtering accounts for large-scale spatial structures but also for intermediate and especially small scale structures and by doing so it is efficient in accounting for spatial autocorrelation (though may introduce some overfitting). This could mean in your context that you first do your RDA with the environmental predictors you are interested in, then use the residuals as a response in an analysis with spatial filters/PCNM as predictors and select those that significantly explain residual variation. Lastly add those selected to the set of environmental predictors. You might then need to reduce environmental predictors again. Variable selection procedure in such cases is to my knowledge not sufficiently solved, yet. However, before accounting for spatial autocorrelation (SAC): Are you sure that your data is affected by this? Dis you test residual autocorrelation structure? If there is not SAC, there is no need to correct... HTH Ingolf Bini, L.M., Diniz-Filho, J.A.F., Rangel, T. et al. 2009. Coefficient shifts in geographical ecology: an empirical evaluation of spatial and non-spatial regression. Ecography 32: 193-204. doi: 10./j.1600-0587.2009.05717.x Diniz-Filho, J.A.F., Bini, L.M., Hawkins, B.A. 2003. Spatial autocorrelation and red herrings in geographical ecology. Global Ecology and Biogeography 12: 53-64. Dray, S., Legendre, P., Peres-Neto, P.R. 2006. Spatial modelling: a comprehensive framework for principal coordinate analysis of neighbour matrices (PCNM). Ecological Modelling 196: 483-493. Am 10.07.2013 21:38, schrieb stephen sefick: > Jari, > > Thank you for the quick reply. Maybe I should use something like PCNM > first with the lat/long data to then use in the rda? I really appreciate > all of your help. Are there anyother/better ways to account for spatial > autocorrelation. I guess I need to show that spatial autocorellation > exists and then if it does account for it? Any reading etc. would be > greatly appreciated. I appreciate all of the help. > kindest regards, > > Stephen > > P.S. I will let you know about the stepwise selection and scope argument > > > On Wed, Jul 10, 2013 at 2:28 PM, Jari Oksanen wrote: > >> On 10/07/2013, at 21:00 PM, Stephen Sefick wrote: >> >>> Hello all, >>> >>> I would like to run this by everyone and maybe get some hints as to what >> R functions I could use for this. Ok, so I have macroinvertebrate >> assemblage data from across the SE. I would like to control for geographic >> distance (lat/long), Watershed area, and year before submitting these data >> to an RDA with the rest of the environmental data using a variable >> selection technique. >>> Does it make sense to detrend the data using a mlm on hellinger >> transfomed abundances with the above env variables as regressors and then >> submit the residuals to rda with the rest of the env variables I am >> interested in? >> >> >> Stephen, >> >> If you happen to use vegan functions for forward selection, please note >> that they all (should) take a scope argument that can (should) be a list of >> lower and upper scopes. Put your controlled variables (distance???, >> watershed area, year) in the lower scope and these plus other candidate >> variables in the upper scope, and there you go. I have used "should", >> because I have rarely used these functions myself, and I'm not sure if >> lower scope really is implemented in all, but is *should* be: file a bug >> report if this fails. >> >> I have no idea how to have distance RDA. Well, I have ideas, but none that >> I have are very good. >> >> Using separate mlm and modelling residuals will not work quite correctly, >> because that ignores correlations between groups of variables. Vegan >> functions do not ignore those. >> >> Cheers, Jari Oksanen >> -- >> Jari Oksanen, Dept Biology, Univ Oulu, 90014 Finland >> jari.oksa...@oulu.fi, Ph. +358 400 408593, http://cc.oulu.fi/~jarioksa >> >> >> >> >> >> > [[alternative HTML version dele
Re: [R-sig-eco] forward selection RDA after controlling for constraints
Jari, Thank you for the quick reply. Maybe I should use something like PCNM first with the lat/long data to then use in the rda? I really appreciate all of your help. Are there anyother/better ways to account for spatial autocorrelation. I guess I need to show that spatial autocorellation exists and then if it does account for it? Any reading etc. would be greatly appreciated. I appreciate all of the help. kindest regards, Stephen P.S. I will let you know about the stepwise selection and scope argument On Wed, Jul 10, 2013 at 2:28 PM, Jari Oksanen wrote: > > On 10/07/2013, at 21:00 PM, Stephen Sefick wrote: > > > Hello all, > > > > I would like to run this by everyone and maybe get some hints as to what > R functions I could use for this. Ok, so I have macroinvertebrate > assemblage data from across the SE. I would like to control for geographic > distance (lat/long), Watershed area, and year before submitting these data > to an RDA with the rest of the environmental data using a variable > selection technique. > > > > Does it make sense to detrend the data using a mlm on hellinger > transfomed abundances with the above env variables as regressors and then > submit the residuals to rda with the rest of the env variables I am > interested in? > > > Stephen, > > If you happen to use vegan functions for forward selection, please note > that they all (should) take a scope argument that can (should) be a list of > lower and upper scopes. Put your controlled variables (distance???, > watershed area, year) in the lower scope and these plus other candidate > variables in the upper scope, and there you go. I have used "should", > because I have rarely used these functions myself, and I'm not sure if > lower scope really is implemented in all, but is *should* be: file a bug > report if this fails. > > I have no idea how to have distance RDA. Well, I have ideas, but none that > I have are very good. > > Using separate mlm and modelling residuals will not work quite correctly, > because that ignores correlations between groups of variables. Vegan > functions do not ignore those. > > Cheers, Jari Oksanen > -- > Jari Oksanen, Dept Biology, Univ Oulu, 90014 Finland > jari.oksa...@oulu.fi, Ph. +358 400 408593, http://cc.oulu.fi/~jarioksa > > > > > > [[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] forward selection RDA after controlling for constraints
On 10/07/2013, at 21:00 PM, Stephen Sefick wrote: > Hello all, > > I would like to run this by everyone and maybe get some hints as to what R > functions I could use for this. Ok, so I have macroinvertebrate assemblage > data from across the SE. I would like to control for geographic distance > (lat/long), Watershed area, and year before submitting these data to an RDA > with the rest of the environmental data using a variable selection technique. > > Does it make sense to detrend the data using a mlm on hellinger transfomed > abundances with the above env variables as regressors and then submit the > residuals to rda with the rest of the env variables I am interested in? Stephen, If you happen to use vegan functions for forward selection, please note that they all (should) take a scope argument that can (should) be a list of lower and upper scopes. Put your controlled variables (distance???, watershed area, year) in the lower scope and these plus other candidate variables in the upper scope, and there you go. I have used "should", because I have rarely used these functions myself, and I'm not sure if lower scope really is implemented in all, but is *should* be: file a bug report if this fails. I have no idea how to have distance RDA. Well, I have ideas, but none that I have are very good. Using separate mlm and modelling residuals will not work quite correctly, because that ignores correlations between groups of variables. Vegan functions do not ignore those. Cheers, Jari Oksanen -- Jari Oksanen, Dept Biology, Univ Oulu, 90014 Finland jari.oksa...@oulu.fi, Ph. +358 400 408593, http://cc.oulu.fi/~jarioksa ___ R-sig-ecology mailing list R-sig-ecology@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-ecology
[R-sig-eco] forward selection RDA after controlling for constraints
Hello all, I would like to run this by everyone and maybe get some hints as to what R functions I could use for this. Ok, so I have macroinvertebrate assemblage data from across the SE. I would like to control for geographic distance (lat/long), Watershed area, and year before submitting these data to an RDA with the rest of the environmental data using a variable selection technique. Does it make sense to detrend the data using a mlm on hellinger transfomed abundances with the above env variables as regressors and then submit the residuals to rda with the rest of the env variables I am interested in? Many thanks for all of the help. kindest regards, -- Stephen Sefick ** Auburn University Biological Sciences 331 Funchess Hall Auburn, Alabama 36849 ** sas0...@auburn.edu http://www.auburn.edu/~sas0025 ** Let's not spend our time and resources thinking about things that are so little or so large that all they really do for us is puff us up and make us feel like gods. We are mammals, and have not exhausted the annoying little problems of being mammals. -K. Mullis "A big computer, a complex algorithm and a long time does not equal science." -Robert Gentleman ___ R-sig-ecology mailing list R-sig-ecology@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-ecology