Re: [R-sig-phylo] compar.ou
Dear all, Following up this conversation; Using OUwie, i get reasonable values for theta under several models for all my traits, but one. In this case, this is a trait which can only take positive values and i get negative theta values for all the different models, including BM and OU1. the standard error is not big enough to consider my theta value to be around zero. (for example: theta = -2.279, se= 0.209). Any idea of what could be causing this and if it is possible to force OUwie to estimate theta within certain boundaries? Thank you, Sandra. 2013/10/29 Marguerite Butler mbutler...@gmail.com Hi Sandra and others, You can also assess confidence using parametric bootstrap, a procedure which we generally recommend for all users. ouch has built-in facilities to do so (the bootstrap() and simulate() functions in addition to update() ). I think there are examples in my tutorial. If not, let me know and I can send you another. Marguerite On Oct 29, 2013, at 6:08 AM, Cecile Ane a...@stat.wisc.edu wrote: FYI, we have some theory to explain why alpha has large standard errors and in which conditions. As Brian says, it comes with a flat likelihood with respect with alpha. http://dx.doi.org/10.1214/13-AOS1105 or http://www.stat.wisc.edu/~ane/publis/2013HoAne_AoS.pdf On 10/29/2013 09:46 AM, Brian O'Meara wrote: In at least the OUwie paper we spent a lot of time doing simulations to determine this empirically (this may have been examined in other papers, too, though none come to mind). Alpha can be estimated, but sometimes with scarily large standard errors (but not always). This property should hold for related methods (it's a property of the likelihood surface for these models). You can just plot the alpha values vs likelihood to get a sense of this surface for your dataset (ideally, estimating the other parameters for each fixed alpha, which is possible, though the kludge of holding the other parameters at their MLE and just varying alpha will give you a sense of the surface, though it's a bit non-conservative). We had code in OUwie to do a contour plot of the likelihood surface but took it out (a small licensing difference between akima (university license) and OUwie (GPL) led to OUwie being pulled from CRAN by the CRAN maintainers until we resolved the difference). Brian ___ Brian O'Meara Assistant Professor Dept. of Ecology Evolutionary Biology U. of Tennessee, Knoxville http://www.brianomeara.info Students wanted: Applications due Dec. 15, annually Postdoc collaborators wanted: Check NIMBioS' website Calendar: http://www.brianomeara.info/calendars/omeara On Tue, Oct 29, 2013 at 10:21 AM, sandra goutte gou...@mnhn.fr wrote: Thank you Marguerite. Looking at OUwie and OUCH/SLOUCH, i see that alpha is estimated along the other parameters, whereas in Hansen 1997 and other papers it is suggested that this would lead to very large standard errors. Is that problem resolved in these functions? Best, Sandra. 2013/10/26 Marguerite Butler mbutler...@gmail.com Dear Sandra, You might also want to look at the papers that go along with slouch, ouch, and ouwie. Here are some pdfs, along with some tutorials. On my Rcompstart, the ouch stuff starts around page 165 Let me know if you need help with ouch. Marguerite On Oct 24, 2013, at 7:03 AM, sandra goutte gou...@mnhn.fr wrote: Dear list, My aim is to compare the fit of models for which *theta* is allowed to change at different nodes (different combinations of 1 ,2 or 3 nodes). I don't really understand the calculation of the deviance, but if i'm not mistaken the difference between the deviances of 2 models follows a *chi*square distribution with the difference of parameters in the models as degree of freedom. Now, how to compare two models with the same number of optimum changes but at different nodes? I am having a hard time understanding the details of the function with regards to Hansen's (1997) paper. What is the relationship between the deviance and the RSS from Hansen (1997)? Is it possible to calculate the RSS from the output of compar.ou ? I would appreciate any advice or insight on the subject! Sandra. -- PhD Student Muséum National d'Histoire Naturelle Département Systématique et Évolution USM 601 / UMR 7205 Origine, Structure et Évolution de la Biodiversité Reptiles Amphibiens - Case Postale 30 25 rue Cuvier F-75005 Paris Tel : +33 (0) 1 40 79 34 90 Mobile: +33 (0) 6 79 20 29 99 [[alternative HTML version deleted]] ___ R-sig-phylo mailing list - R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/
Re: [R-sig-phylo] compar.ou
Often if you have a trait constrained to be positive, it's appropriate to log-transform it, which has the happy effect of making the theta for the trait when converted back to a non-log scale constrained to be positive (as well as probably being more appropriate for how your trait evolves). You can have theta outside the range of observed values, but this should only happen in something like OU1 if your tree is not ultrametric (ultrametric = equal root to tip length, as with a chronogram of extant species), and even then is unlikely (basically, it would only happen if you have something like a trend). You could do a constrained search by wrapping OUwie.fixed() in an optimizer where you return a really bad, but finite, likelihood if you exceed one of your bounds (or use an optimizer that allows setting bounds), or, perhaps easier, modify the OUwie() function to return a bad likelihood for particular values of theta. But my first suggestion is to think about whether your data are better log-transformed (i.e, under Brownian motion, is an increase or decrease of 10% of the present value equally likely regardless of trait value (in which case, log transformation makes sense)). You also may want to check that the branch lengths you are using make sense for your problem (OU, BM, etc. models are largely (not entirely) about transforming branch lengths on trees, so you want to make sure they are sensible for your question to start with: do you care about rates with time units or amount of molecular change?). Best, Brian ___ Brian O'Meara Assistant Professor Dept. of Ecology Evolutionary Biology U. of Tennessee, Knoxville http://www.brianomeara.info Students wanted: Applications due Dec. 15, annually Postdoc collaborators wanted: Check NIMBioS' website Calendar: http://www.brianomeara.info/calendars/omeara On Tue, Nov 12, 2013 at 6:39 AM, sandra goutte gou...@mnhn.fr wrote: Dear all, Following up this conversation; Using OUwie, i get reasonable values for theta under several models for all my traits, but one. In this case, this is a trait which can only take positive values and i get negative theta values for all the different models, including BM and OU1. the standard error is not big enough to consider my theta value to be around zero. (for example: theta = -2.279, se= 0.209). Any idea of what could be causing this and if it is possible to force OUwie to estimate theta within certain boundaries? Thank you, Sandra. 2013/10/29 Marguerite Butler mbutler...@gmail.com Hi Sandra and others, You can also assess confidence using parametric bootstrap, a procedure which we generally recommend for all users. ouch has built-in facilities to do so (the bootstrap() and simulate() functions in addition to update() ). I think there are examples in my tutorial. If not, let me know and I can send you another. Marguerite On Oct 29, 2013, at 6:08 AM, Cecile Ane a...@stat.wisc.edu wrote: FYI, we have some theory to explain why alpha has large standard errors and in which conditions. As Brian says, it comes with a flat likelihood with respect with alpha. http://dx.doi.org/10.1214/13-AOS1105 or http://www.stat.wisc.edu/~ane/publis/2013HoAne_AoS.pdf On 10/29/2013 09:46 AM, Brian O'Meara wrote: In at least the OUwie paper we spent a lot of time doing simulations to determine this empirically (this may have been examined in other papers, too, though none come to mind). Alpha can be estimated, but sometimes with scarily large standard errors (but not always). This property should hold for related methods (it's a property of the likelihood surface for these models). You can just plot the alpha values vs likelihood to get a sense of this surface for your dataset (ideally, estimating the other parameters for each fixed alpha, which is possible, though the kludge of holding the other parameters at their MLE and just varying alpha will give you a sense of the surface, though it's a bit non-conservative). We had code in OUwie to do a contour plot of the likelihood surface but took it out (a small licensing difference between akima (university license) and OUwie (GPL) led to OUwie being pulled from CRAN by the CRAN maintainers until we resolved the difference). Brian ___ Brian O'Meara Assistant Professor Dept. of Ecology Evolutionary Biology U. of Tennessee, Knoxville http://www.brianomeara.info Students wanted: Applications due Dec. 15, annually Postdoc collaborators wanted: Check NIMBioS' website Calendar: http://www.brianomeara.info/calendars/omeara On Tue, Oct 29, 2013 at 10:21 AM, sandra goutte gou...@mnhn.fr wrote: Thank you Marguerite. Looking at OUwie and OUCH/SLOUCH, i see that alpha is estimated along the other parameters, whereas in Hansen 1997 and other papers
Re: [R-sig-phylo] compar.ou
Thank you Marguerite. Looking at OUwie and OUCH/SLOUCH, i see that alpha is estimated along the other parameters, whereas in Hansen 1997 and other papers it is suggested that this would lead to very large standard errors. Is that problem resolved in these functions? Best, Sandra. 2013/10/26 Marguerite Butler mbutler...@gmail.com Dear Sandra, You might also want to look at the papers that go along with slouch, ouch, and ouwie. Here are some pdfs, along with some tutorials. On my Rcompstart, the ouch stuff starts around page 165 Let me know if you need help with ouch. Marguerite On Oct 24, 2013, at 7:03 AM, sandra goutte gou...@mnhn.fr wrote: Dear list, My aim is to compare the fit of models for which *theta* is allowed to change at different nodes (different combinations of 1 ,2 or 3 nodes). I don't really understand the calculation of the deviance, but if i'm not mistaken the difference between the deviances of 2 models follows a *chi*square distribution with the difference of parameters in the models as degree of freedom. Now, how to compare two models with the same number of optimum changes but at different nodes? I am having a hard time understanding the details of the function with regards to Hansen's (1997) paper. What is the relationship between the deviance and the RSS from Hansen (1997)? Is it possible to calculate the RSS from the output of compar.ou ? I would appreciate any advice or insight on the subject! Sandra. -- PhD Student Muséum National d'Histoire Naturelle Département Systématique et Évolution USM 601 / UMR 7205 Origine, Structure et Évolution de la Biodiversité Reptiles Amphibiens - Case Postale 30 25 rue Cuvier F-75005 Paris Tel : +33 (0) 1 40 79 34 90 Mobile: +33 (0) 6 79 20 29 99 [[alternative HTML version deleted]] ___ R-sig-phylo mailing list - R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/ Marguerite A. Butler Associate Professor Department of Biology University of Hawaii 2538 McCarthy Mall, Edmondson Hall 216 Honolulu HI 96822 Office: 808-956-4713 Dept: 808-956-8617 Lab: 808-956-5867 FAX: 808-956-9812 http://www.hawaii.edu/zoology/faculty/butler.html http://www2.hawaii.edu/~mbutler http://www.hawaii.edu/zoology/ -- PhD Student Muséum National d'Histoire Naturelle Département Systématique et Évolution USM 601 / UMR 7205 Origine, Structure et Évolution de la Biodiversité Reptiles Amphibiens - Case Postale 30 25 rue Cuvier F-75005 Paris Tel : +33 (0) 1 40 79 34 90 Mobile: +33 (0) 6 79 20 29 99 [[alternative HTML version deleted]] ___ R-sig-phylo mailing list - R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/
Re: [R-sig-phylo] compar.ou
In at least the OUwie paper we spent a lot of time doing simulations to determine this empirically (this may have been examined in other papers, too, though none come to mind). Alpha can be estimated, but sometimes with scarily large standard errors (but not always). This property should hold for related methods (it's a property of the likelihood surface for these models). You can just plot the alpha values vs likelihood to get a sense of this surface for your dataset (ideally, estimating the other parameters for each fixed alpha, which is possible, though the kludge of holding the other parameters at their MLE and just varying alpha will give you a sense of the surface, though it's a bit non-conservative). We had code in OUwie to do a contour plot of the likelihood surface but took it out (a small licensing difference between akima (university license) and OUwie (GPL) led to OUwie being pulled from CRAN by the CRAN maintainers until we resolved the difference). Brian ___ Brian O'Meara Assistant Professor Dept. of Ecology Evolutionary Biology U. of Tennessee, Knoxville http://www.brianomeara.info Students wanted: Applications due Dec. 15, annually Postdoc collaborators wanted: Check NIMBioS' website Calendar: http://www.brianomeara.info/calendars/omeara On Tue, Oct 29, 2013 at 10:21 AM, sandra goutte gou...@mnhn.fr wrote: Thank you Marguerite. Looking at OUwie and OUCH/SLOUCH, i see that alpha is estimated along the other parameters, whereas in Hansen 1997 and other papers it is suggested that this would lead to very large standard errors. Is that problem resolved in these functions? Best, Sandra. 2013/10/26 Marguerite Butler mbutler...@gmail.com Dear Sandra, You might also want to look at the papers that go along with slouch, ouch, and ouwie. Here are some pdfs, along with some tutorials. On my Rcompstart, the ouch stuff starts around page 165 Let me know if you need help with ouch. Marguerite On Oct 24, 2013, at 7:03 AM, sandra goutte gou...@mnhn.fr wrote: Dear list, My aim is to compare the fit of models for which *theta* is allowed to change at different nodes (different combinations of 1 ,2 or 3 nodes). I don't really understand the calculation of the deviance, but if i'm not mistaken the difference between the deviances of 2 models follows a *chi*square distribution with the difference of parameters in the models as degree of freedom. Now, how to compare two models with the same number of optimum changes but at different nodes? I am having a hard time understanding the details of the function with regards to Hansen's (1997) paper. What is the relationship between the deviance and the RSS from Hansen (1997)? Is it possible to calculate the RSS from the output of compar.ou ? I would appreciate any advice or insight on the subject! Sandra. -- PhD Student Muséum National d'Histoire Naturelle Département Systématique et Évolution USM 601 / UMR 7205 Origine, Structure et Évolution de la Biodiversité Reptiles Amphibiens - Case Postale 30 25 rue Cuvier F-75005 Paris Tel : +33 (0) 1 40 79 34 90 Mobile: +33 (0) 6 79 20 29 99 [[alternative HTML version deleted]] ___ R-sig-phylo mailing list - R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/ Marguerite A. Butler Associate Professor Department of Biology University of Hawaii 2538 McCarthy Mall, Edmondson Hall 216 Honolulu HI 96822 Office: 808-956-4713 Dept: 808-956-8617 Lab: 808-956-5867 FAX: 808-956-9812 http://www.hawaii.edu/zoology/faculty/butler.html http://www2.hawaii.edu/~mbutler http://www.hawaii.edu/zoology/ -- PhD Student Muséum National d'Histoire Naturelle Département Systématique et Évolution USM 601 / UMR 7205 Origine, Structure et Évolution de la Biodiversité Reptiles Amphibiens - Case Postale 30 25 rue Cuvier F-75005 Paris Tel : +33 (0) 1 40 79 34 90 Mobile: +33 (0) 6 79 20 29 99 [[alternative HTML version deleted]] ___ R-sig-phylo mailing list - R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/ [[alternative HTML version deleted]] ___ R-sig-phylo mailing list - R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/
Re: [R-sig-phylo] compar.ou
FYI, we have some theory to explain why alpha has large standard errors and in which conditions. As Brian says, it comes with a flat likelihood with respect with alpha. http://dx.doi.org/10.1214/13-AOS1105 or http://www.stat.wisc.edu/~ane/publis/2013HoAne_AoS.pdf On 10/29/2013 09:46 AM, Brian O'Meara wrote: In at least the OUwie paper we spent a lot of time doing simulations to determine this empirically (this may have been examined in other papers, too, though none come to mind). Alpha can be estimated, but sometimes with scarily large standard errors (but not always). This property should hold for related methods (it's a property of the likelihood surface for these models). You can just plot the alpha values vs likelihood to get a sense of this surface for your dataset (ideally, estimating the other parameters for each fixed alpha, which is possible, though the kludge of holding the other parameters at their MLE and just varying alpha will give you a sense of the surface, though it's a bit non-conservative). We had code in OUwie to do a contour plot of the likelihood surface but took it out (a small licensing difference between akima (university license) and OUwie (GPL) led to OUwie being pulled from CRAN by the CRAN maintainers until we resolved the difference). Brian ___ Brian O'Meara Assistant Professor Dept. of Ecology Evolutionary Biology U. of Tennessee, Knoxville http://www.brianomeara.info Students wanted: Applications due Dec. 15, annually Postdoc collaborators wanted: Check NIMBioS' website Calendar: http://www.brianomeara.info/calendars/omeara On Tue, Oct 29, 2013 at 10:21 AM, sandra goutte gou...@mnhn.fr wrote: Thank you Marguerite. Looking at OUwie and OUCH/SLOUCH, i see that alpha is estimated along the other parameters, whereas in Hansen 1997 and other papers it is suggested that this would lead to very large standard errors. Is that problem resolved in these functions? Best, Sandra. 2013/10/26 Marguerite Butler mbutler...@gmail.com Dear Sandra, You might also want to look at the papers that go along with slouch, ouch, and ouwie. Here are some pdfs, along with some tutorials. On my Rcompstart, the ouch stuff starts around page 165 Let me know if you need help with ouch. Marguerite On Oct 24, 2013, at 7:03 AM, sandra goutte gou...@mnhn.fr wrote: Dear list, My aim is to compare the fit of models for which *theta* is allowed to change at different nodes (different combinations of 1 ,2 or 3 nodes). I don't really understand the calculation of the deviance, but if i'm not mistaken the difference between the deviances of 2 models follows a *chi*square distribution with the difference of parameters in the models as degree of freedom. Now, how to compare two models with the same number of optimum changes but at different nodes? I am having a hard time understanding the details of the function with regards to Hansen's (1997) paper. What is the relationship between the deviance and the RSS from Hansen (1997)? Is it possible to calculate the RSS from the output of compar.ou ? I would appreciate any advice or insight on the subject! Sandra. -- PhD Student Muséum National d'Histoire Naturelle Département Systématique et Évolution USM 601 / UMR 7205 Origine, Structure et Évolution de la Biodiversité Reptiles Amphibiens - Case Postale 30 25 rue Cuvier F-75005 Paris Tel : +33 (0) 1 40 79 34 90 Mobile: +33 (0) 6 79 20 29 99 [[alternative HTML version deleted]] ___ R-sig-phylo mailing list - R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/ Marguerite A. Butler Associate Professor Department of Biology University of Hawaii 2538 McCarthy Mall, Edmondson Hall 216 Honolulu HI 96822 Office: 808-956-4713 Dept: 808-956-8617 Lab: 808-956-5867 FAX: 808-956-9812 http://www.hawaii.edu/zoology/faculty/butler.html http://www2.hawaii.edu/~mbutler http://www.hawaii.edu/zoology/ -- Cecile Ane Departments of Statistics and of Botany University of Wisconsin - Madison www.stat.wisc.edu/~ane/ CALS statistical consulting lab: www.cals.wisc.edu/calslab/stat_consulting.php ___ R-sig-phylo mailing list - R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/
Re: [R-sig-phylo] compar.ou
Hi Sandra and others, You can also assess confidence using parametric bootstrap, a procedure which we generally recommend for all users. ouch has built-in facilities to do so (the bootstrap() and simulate() functions in addition to update() ). I think there are examples in my tutorial. If not, let me know and I can send you another. Marguerite On Oct 29, 2013, at 6:08 AM, Cecile Ane a...@stat.wisc.edu wrote: FYI, we have some theory to explain why alpha has large standard errors and in which conditions. As Brian says, it comes with a flat likelihood with respect with alpha. http://dx.doi.org/10.1214/13-AOS1105 or http://www.stat.wisc.edu/~ane/publis/2013HoAne_AoS.pdf On 10/29/2013 09:46 AM, Brian O'Meara wrote: In at least the OUwie paper we spent a lot of time doing simulations to determine this empirically (this may have been examined in other papers, too, though none come to mind). Alpha can be estimated, but sometimes with scarily large standard errors (but not always). This property should hold for related methods (it's a property of the likelihood surface for these models). You can just plot the alpha values vs likelihood to get a sense of this surface for your dataset (ideally, estimating the other parameters for each fixed alpha, which is possible, though the kludge of holding the other parameters at their MLE and just varying alpha will give you a sense of the surface, though it's a bit non-conservative). We had code in OUwie to do a contour plot of the likelihood surface but took it out (a small licensing difference between akima (university license) and OUwie (GPL) led to OUwie being pulled from CRAN by the CRAN maintainers until we resolved the difference). Brian ___ Brian O'Meara Assistant Professor Dept. of Ecology Evolutionary Biology U. of Tennessee, Knoxville http://www.brianomeara.info Students wanted: Applications due Dec. 15, annually Postdoc collaborators wanted: Check NIMBioS' website Calendar: http://www.brianomeara.info/calendars/omeara On Tue, Oct 29, 2013 at 10:21 AM, sandra goutte gou...@mnhn.fr wrote: Thank you Marguerite. Looking at OUwie and OUCH/SLOUCH, i see that alpha is estimated along the other parameters, whereas in Hansen 1997 and other papers it is suggested that this would lead to very large standard errors. Is that problem resolved in these functions? Best, Sandra. 2013/10/26 Marguerite Butler mbutler...@gmail.com Dear Sandra, You might also want to look at the papers that go along with slouch, ouch, and ouwie. Here are some pdfs, along with some tutorials. On my Rcompstart, the ouch stuff starts around page 165 Let me know if you need help with ouch. Marguerite On Oct 24, 2013, at 7:03 AM, sandra goutte gou...@mnhn.fr wrote: Dear list, My aim is to compare the fit of models for which *theta* is allowed to change at different nodes (different combinations of 1 ,2 or 3 nodes). I don't really understand the calculation of the deviance, but if i'm not mistaken the difference between the deviances of 2 models follows a *chi*square distribution with the difference of parameters in the models as degree of freedom. Now, how to compare two models with the same number of optimum changes but at different nodes? I am having a hard time understanding the details of the function with regards to Hansen's (1997) paper. What is the relationship between the deviance and the RSS from Hansen (1997)? Is it possible to calculate the RSS from the output of compar.ou ? I would appreciate any advice or insight on the subject! Sandra. -- PhD Student Muséum National d'Histoire Naturelle Département Systématique et Évolution USM 601 / UMR 7205 Origine, Structure et Évolution de la Biodiversité Reptiles Amphibiens - Case Postale 30 25 rue Cuvier F-75005 Paris Tel : +33 (0) 1 40 79 34 90 Mobile: +33 (0) 6 79 20 29 99 [[alternative HTML version deleted]] ___ R-sig-phylo mailing list - R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/ Marguerite A. Butler Associate Professor Department of Biology University of Hawaii 2538 McCarthy Mall, Edmondson Hall 216 Honolulu HI 96822 Office: 808-956-4713 Dept: 808-956-8617 Lab: 808-956-5867 FAX: 808-956-9812 http://www.hawaii.edu/zoology/faculty/butler.html http://www2.hawaii.edu/~mbutler http://www.hawaii.edu/zoology/ -- Cecile Ane Departments of Statistics and of Botany University of Wisconsin - Madison www.stat.wisc.edu/~ane/ CALS statistical consulting lab: www.cals.wisc.edu/calslab/stat_consulting.php ___ R-sig-phylo mailing list - R-sig-phylo@r-project.org
Re: [R-sig-phylo] compar.ou (sandra goutte)
Hi Sandra, I don't know about compar.ou but my mvSLOUCH package (returns RSS and allows you to have measurement error and missing data) and Butler King's ouch package return you a lot of stastics that you can use for model comparison. Maybe in your case the AIC.C would be more appropriate? Best wishes Krzysztof Bartoszek ___ R-sig-phylo mailing list - R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/
Re: [R-sig-phylo] compar.ou (sandra goutte)
You can also use OUwie for a variety of models (OU with different means and/or different variances and/or different attraction values), but it returns AIC and lnL scores but not RSS. It sounds, though, that mvSLOUCH might be the best option in your case. Best, Brian ___ Brian O'Meara Assistant Professor Dept. of Ecology Evolutionary Biology U. of Tennessee, Knoxville http://www.brianomeara.info Students wanted: Applications due Dec. 15, annually Postdoc collaborators wanted: Check NIMBioS' website Calendar: http://www.brianomeara.info/calendars/omeara On Fri, Oct 25, 2013 at 8:47 AM, Krzysztof Bartoszek krz...@chalmers.sewrote: Hi Sandra, I don't know about compar.ou but my mvSLOUCH package (returns RSS and allows you to have measurement error and missing data) and Butler King's ouch package return you a lot of stastics that you can use for model comparison. Maybe in your case the AIC.C would be more appropriate? Best wishes Krzysztof Bartoszek ___ R-sig-phylo mailing list - R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/ [[alternative HTML version deleted]] ___ R-sig-phylo mailing list - R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/
Re: [R-sig-phylo] compar.ou (sandra goutte)
Thank you both for your answers, i will look into those packages. Best, Sandra. 2013/10/25 Brian O'Meara bome...@utk.edu You can also use OUwie for a variety of models (OU with different means and/or different variances and/or different attraction values), but it returns AIC and lnL scores but not RSS. It sounds, though, that mvSLOUCH might be the best option in your case. Best, Brian ___ Brian O'Meara Assistant Professor Dept. of Ecology Evolutionary Biology U. of Tennessee, Knoxville http://www.brianomeara.info Students wanted: Applications due Dec. 15, annually Postdoc collaborators wanted: Check NIMBioS' website Calendar: http://www.brianomeara.info/calendars/omeara On Fri, Oct 25, 2013 at 8:47 AM, Krzysztof Bartoszek krz...@chalmers.se wrote: Hi Sandra, I don't know about compar.ou but my mvSLOUCH package (returns RSS and allows you to have measurement error and missing data) and Butler King's ouch package return you a lot of stastics that you can use for model comparison. Maybe in your case the AIC.C would be more appropriate? Best wishes Krzysztof Bartoszek ___ R-sig-phylo mailing list - R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/ [[alternative HTML version deleted]] ___ R-sig-phylo mailing list - R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/ -- PhD Student Muséum National d'Histoire Naturelle Département Systématique et Évolution USM 601 / UMR 7205 Origine, Structure et Évolution de la Biodiversité Reptiles Amphibiens - Case Postale 30 25 rue Cuvier F-75005 Paris Tel : +33 (0) 1 40 79 34 90 Mobile: +33 (0) 6 79 20 29 99 [[alternative HTML version deleted]] ___ R-sig-phylo mailing list - R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/
[R-sig-phylo] compar.ou
Dear list, My aim is to compare the fit of models for which *theta* is allowed to change at different nodes (different combinations of 1 ,2 or 3 nodes). I don't really understand the calculation of the deviance, but if i'm not mistaken the difference between the deviances of 2 models follows a *chi*square distribution with the difference of parameters in the models as degree of freedom. Now, how to compare two models with the same number of optimum changes but at different nodes? I am having a hard time understanding the details of the function with regards to Hansen's (1997) paper. What is the relationship between the deviance and the RSS from Hansen (1997)? Is it possible to calculate the RSS from the output of compar.ou ? I would appreciate any advice or insight on the subject! Sandra. -- PhD Student Muséum National d'Histoire Naturelle Département Systématique et Évolution USM 601 / UMR 7205 Origine, Structure et Évolution de la Biodiversité Reptiles Amphibiens - Case Postale 30 25 rue Cuvier F-75005 Paris Tel : +33 (0) 1 40 79 34 90 Mobile: +33 (0) 6 79 20 29 99 [[alternative HTML version deleted]] ___ R-sig-phylo mailing list - R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/
[R-sig-phylo] compar.ou in ape package - warnings and error messages on bird.orders example dataset
Dear list, When running compar.ou example from APE package documentation (v. 23-11-09), I've got couple of error messages. What might be the reason for them: compar.ou(rnorm(23), bird.orders, alpha = 0.1) $deviance [1] 57.60627 $para estimatestderr sigma2 0.71937096 0.1500408 theta1 -0.04114436 0.1365702 $call compar.ou(x = rnorm(23), phy = bird.orders, alpha = 0.1) attr(,class) [1] compar.ou Warning messages: 1: In log(2 * pi * p[2]) : NaNs produced 2: In nlm(function(p) dev(c(alpha, p)), p = c(1, rep(mean(x), ncol(Wstart))), : NA/Inf replaced by maximum positive value y - c(rnorm(5, 0), rnorm(18, 5)) compar.ou(y, bird.orders, node = -2, alpha = .1) Error in dimnames(x) - dn : length of 'dimnames' [2] not equal to array extent Thanks a lot for any clue, Best regards, Anna ___ R-sig-phylo mailing list R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo