Xochitl CORMON
2015-Jan-28 14:08 UTC
[R] Prediction of response after glm on whitened data
Hi all,
Here is a description of my case. I am sorry if my question is also
statistic related but it is difficult to disentangle. I will however try
to make it only R applied.
My response is a growth constant "k" and my descriptor is prey biomass
"NP" and time series is of 21 years.
I applied a gaussiam GLM (or LM) to this question. After the regression
I tested the residuals for autocorrelation using acf(). Because
autocorrelation was significant I decided to whiten my data using
{car}dwt() in order to obtain rho (an estimation of my correlation) and
then applying the following to my data in order to remove autocorrelation:
kw_i = k_i - rho * k_i-1
NPw_i = NPw_i - rho * NPw_i-1
(method from Jonathan Taylor,
http://statweb.stanford.edu/~jtaylo/courses/stats191/correlated_errors.html).
After that I fitted a model on this whitened data (kw_i ~ NPw_i),
realised an F-test and obtained classical results such as deviance
explained, pvalues and of course the intercept and coefficient of the
last regression. However doing that and coming to prediction using
predict() I can only obtained predictions of deltaK (kw_i) in function
of deltaNP (NPw_i) but I am actually interested in being able to predict
k in function of NP...
Is there a solution to predict directly k and its associated variance
using R without having to detail in the script all the mathematical
process necessary to come back to something like k_i = mu + rho * k_i-1
+ beta(NPw_i - rho * NPw_i-1) + epsilon
with mu being the intercept, beta the regression coefficient and epsilon
the error,
?
Thank you for your help,
Best,
Xochitl C.
--
<>< <>< <>< <><
Xochitl CORMON
+33 (0)3 21 99 56 84
Doctorante en ?cologie marine et science halieutique
PhD student in marine ecology and fishery science
<>< <>< <>< <><
IFREMER
Centre Manche Mer du Nord
150 quai Gambetta
62200 Boulogne-sur-Mer
<>< <>< <>< <><
ONKELINX, Thierry
2015-Jan-28 15:19 UTC
[R] Prediction of response after glm on whitened data
Dear Xochitl,
Have a look at gls() from the nlme package. It allows you to fit auto correlated
errors.
gls(k ~ NPw, correlation = corAR1(form = ~ Time))
Best regards,
ir. Thierry Onkelinx
Instituut voor natuur- en bosonderzoek / Research Institute for Nature and
Forest
team Biometrie & Kwaliteitszorg / team Biometrics & Quality Assurance
Kliniekstraat 25
1070 Anderlecht
Belgium
+ 32 2 525 02 51
+ 32 54 43 61 85
Thierry.Onkelinx at inbo.be
www.inbo.be
To call in the statistician after the experiment is done may be no more than
asking him to perform a post-mortem examination: he may be able to say what the
experiment died of.
~ Sir Ronald Aylmer Fisher
The plural of anecdote is not data.
~ Roger Brinner
The combination of some data and an aching desire for an answer does not ensure
that a reasonable answer can be extracted from a given body of data.
~ John Tukey
-----Oorspronkelijk bericht-----
Van: R-help [mailto:r-help-bounces at r-project.org] Namens Xochitl CORMON
Verzonden: woensdag 28 januari 2015 15:09
Aan: Rlist; Rlist
Onderwerp: [R] Prediction of response after glm on whitened data
Hi all,
Here is a description of my case. I am sorry if my question is also statistic
related but it is difficult to disentangle. I will however try to make it only R
applied.
My response is a growth constant "k" and my descriptor is prey biomass
"NP" and time series is of 21 years.
I applied a gaussiam GLM (or LM) to this question. After the regression I tested
the residuals for autocorrelation using acf(). Because autocorrelation was
significant I decided to whiten my data using
{car}dwt() in order to obtain rho (an estimation of my correlation) and then
applying the following to my data in order to remove autocorrelation:
kw_i = k_i - rho * k_i-1
NPw_i = NPw_i - rho * NPw_i-1
(method from Jonathan Taylor,
http://statweb.stanford.edu/~jtaylo/courses/stats191/correlated_errors.html).
After that I fitted a model on this whitened data (kw_i ~ NPw_i), realised an
F-test and obtained classical results such as deviance explained, pvalues and of
course the intercept and coefficient of the last regression. However doing that
and coming to prediction using
predict() I can only obtained predictions of deltaK (kw_i) in function of
deltaNP (NPw_i) but I am actually interested in being able to predict k in
function of NP...
Is there a solution to predict directly k and its associated variance using R
without having to detail in the script all the mathematical process necessary to
come back to something like k_i = mu + rho * k_i-1
+ beta(NPw_i - rho * NPw_i-1) + epsilon
with mu being the intercept, beta the regression coefficient and epsilon the
error, ?
Thank you for your help,
Best,
Xochitl C.
--
<>< <>< <>< <><
Xochitl CORMON
+33 (0)3 21 99 56 84
Doctorante en ?cologie marine et science halieutique PhD student in marine
ecology and fishery science
<>< <>< <>< <><
IFREMER
Centre Manche Mer du Nord
150 quai Gambetta
62200 Boulogne-sur-Mer
<>< <>< <>< <><
______________________________________________
R-help at r-project.org mailing list -- To UNSUBSCRIBE and more, see
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
Disclaimer<https://www.inbo.be/nl/disclaimer-mailberichten-van-het-inbo>
Xochitl CORMON
2015-Jan-29 11:44 UTC
[R] Prediction of response after glm on whitened data
Thanks Thierry for your quick answer. Indeed this simplifies a lot my method so I decided to apply it. However I will be curious to check in which extend the coefficients obtained with the gls function are similar to the ones obtained using glm and whitening. It seems to me thant the method are indeed pretty similar. So if someone knows a function which allows me to predict my response and its associated variance using R after whitening and glm (see original question), I am still eager to know it. Best, Xo <>< <>< <>< <>< Xochitl CORMON +33 (0)3 21 99 56 84 Doctorante en ?cologie marine et science halieutique PhD student in marine ecology and fishery science <>< <>< <>< <>< IFREMER Centre Manche Mer du Nord 150 quai Gambetta 62200 Boulogne-sur-Mer <>< <>< <>< <>< Le 28/01/2015 16:19, ONKELINX, Thierry a ?crit :> Dear Xochitl, > > Have a look at gls() from the nlme package. It allows you to fit auto correlated errors. > > gls(k ~ NPw, correlation = corAR1(form = ~ Time)) > > Best regards, > > ir. Thierry Onkelinx > Instituut voor natuur- en bosonderzoek / Research Institute for Nature and Forest > team Biometrie & Kwaliteitszorg / team Biometrics & Quality Assurance > Kliniekstraat 25 > 1070 Anderlecht > Belgium > + 32 2 525 02 51 > + 32 54 43 61 85 > Thierry.Onkelinx at inbo.be > www.inbo.be > > To call in the statistician after the experiment is done may be no more than asking him to perform a post-mortem examination: he may be able to say what the experiment died of. > ~ Sir Ronald Aylmer Fisher > > The plural of anecdote is not data. > ~ Roger Brinner > > The combination of some data and an aching desire for an answer does not ensure that a reasonable answer can be extracted from a given body of data. > ~ John Tukey > > > -----Oorspronkelijk bericht----- > Van: R-help [mailto:r-help-bounces at r-project.org] Namens Xochitl CORMON > Verzonden: woensdag 28 januari 2015 15:09 > Aan: Rlist; Rlist > Onderwerp: [R] Prediction of response after glm on whitened data > > Hi all, > > Here is a description of my case. I am sorry if my question is also statistic related but it is difficult to disentangle. I will however try to make it only R applied. > > My response is a growth constant "k" and my descriptor is prey biomass "NP" and time series is of 21 years. > > I applied a gaussiam GLM (or LM) to this question. After the regression I tested the residuals for autocorrelation using acf(). Because autocorrelation was significant I decided to whiten my data using > {car}dwt() in order to obtain rho (an estimation of my correlation) and then applying the following to my data in order to remove autocorrelation: > kw_i = k_i - rho * k_i-1 > NPw_i = NPw_i - rho * NPw_i-1 > (method from Jonathan Taylor, > http://statweb.stanford.edu/~jtaylo/courses/stats191/correlated_errors.html). > > After that I fitted a model on this whitened data (kw_i ~ NPw_i), realised an F-test and obtained classical results such as deviance explained, pvalues and of course the intercept and coefficient of the last regression. However doing that and coming to prediction using > predict() I can only obtained predictions of deltaK (kw_i) in function of deltaNP (NPw_i) but I am actually interested in being able to predict k in function of NP... > > Is there a solution to predict directly k and its associated variance using R without having to detail in the script all the mathematical process necessary to come back to something like k_i = mu + rho * k_i-1 > + beta(NPw_i - rho * NPw_i-1) + epsilon > with mu being the intercept, beta the regression coefficient and epsilon the error, ? > > Thank you for your help, > > Best, > > Xochitl C. > > > -- > > <>< <>< <>< <>< > > Xochitl CORMON > +33 (0)3 21 99 56 84 > > Doctorante en ?cologie marine et science halieutique PhD student in marine ecology and fishery science > > <>< <>< <>< <>< > > IFREMER > Centre Manche Mer du Nord > 150 quai Gambetta > 62200 Boulogne-sur-Mer > > <>< <>< <>< <>< > > ______________________________________________ > R-help at r-project.org mailing list -- To UNSUBSCRIBE and more, see https://stat.ethz.ch/mailman/listinfo/r-help > PLEASE do read the posting guide http://www.R-project.org/posting-guide.html > and provide commented, minimal, self-contained, reproducible code. > Disclaimer<https://www.inbo.be/nl/disclaimer-mailberichten-van-het-inbo> >