Hi,
I am using r.series to calculate linear regression slope, offset and coefficient of determination. But any idea how I can get the standard deviation t-value and p-value of the slope?
Cheers,
Paulo
On Tue, Nov 6, 2012 at 9:55 AM, Paulo van Breugel
<p.vanbreugel@gmail.com> wrote:
Hi,
I am using r.series to calculate linear regression slope, offset and
coefficient of determination. But any idea how I can get the standard
deviation t-value and p-value of the slope?
It is not yet implemented... suggestions where to find an existing
C implementation which could be "recycled" (i.e. be GPL compliant)
are welcome.
Markus
OK, thanks for the info.
···
On Wed, Nov 7, 2012 at 9:45 AM, Markus Neteler <neteler@osgeo.org> wrote:
On Tue, Nov 6, 2012 at 9:55 AM, Paulo van Breugel
<p.vanbreugel@gmail.com> wrote:Hi,
I am using r.series to calculate linear regression slope, offset and
coefficient of determination. But any idea how I can get the standard
deviation t-value and p-value of the slope?It is not yet implemented… suggestions where to find an existing
C implementation which could be “recycled” (i.e. be GPL compliant)
are welcome.Markus
I have not clue about C programming, but could there be anything in the Gnu Scientific Library (GSL - http://www.gnu.org/software/gsl/)? See also this question on StackOverflow: http://stackoverflow.com/questions/5503733/getting-p-value-for-linear-regression-in-c-gsl-fit-linear-function-from-gsl-li
Cheers,
Paulo
···
On Wed, Nov 7, 2012 at 9:53 AM, Paulo van Breugel <p.vanbreugel@gmail.com> wrote:
OK, thanks for the info.
On Wed, Nov 7, 2012 at 9:45 AM, Markus Neteler <neteler@osgeo.org> wrote:
On Tue, Nov 6, 2012 at 9:55 AM, Paulo van Breugel
<p.vanbreugel@gmail.com> wrote:Hi,
I am using r.series to calculate linear regression slope, offset and
coefficient of determination. But any idea how I can get the standard
deviation t-value and p-value of the slope?It is not yet implemented… suggestions where to find an existing
C implementation which could be “recycled” (i.e. be GPL compliant)
are welcome.Markus
This is also my urgent need fuction. waiting …
-----原始邮件-----
发件人: “Markus Neteler” neteler@osgeo.org
发送时间: 2012年11月7日 星期三
收件人: “Paulo van Breugel” p.vanbreugel@gmail.com
抄送: “GRASS users email list” grass-user@lists.osgeo.org
主题: Re: [GRASS-user] Calculate p value for regression slope in r.seriesOn Tue, Nov 6, 2012 at 9:55 AM, Paulo van Breugel
p.vanbreugel@gmail.com wrote:Hi,
I am using r.series to calculate linear regression slope, offset and
coefficient of determination. But any idea how I can get the standard
deviation t-value and p-value of the slope?It is not yet implemented… suggestions where to find an existing
C implementation which could be “recycled” (i.e. be GPL compliant)
are welcome.Markus
grass-user mailing list
grass-user@lists.osgeo.org
http://lists.osgeo.org/mailman/listinfo/grass-user
On 06/11/12 09:55, Paulo van Breugel wrote:
Hi,
I am using r.series to calculate linear regression slope, offset and
coefficient of determination. But any idea how I can get the standard
deviation t-value and p-value of the slope?
Try MarkusN's r.regression.series in the addons for grass7. It allows the calculation of t and f values. IIRC, p-value can be calculated from t-value.
Moritz
Good suggestion, and coincidentally I was just looking at it. Am I correct to assume that to mimic the calculation of slope etc in r.series, I would need to create a second time series of ‘dummy’ maps with values 0 for first time period, 1 for second time period, etc.? I did this in R and it gave the same slope and offset as r.series.
Just wondering, if it is implemented in r.regression.series, would it be very difficult to port to r.series? Or do the two things function different internally?
Cheers,
Paulo
···
On Thu, Nov 8, 2012 at 10:55 AM, Moritz Lennert <mlennert@club.worldonline.be> wrote:
On 06/11/12 09:55, Paulo van Breugel wrote:
Hi,
I am using r.series to calculate linear regression slope, offset and
coefficient of determination. But any idea how I can get the standard
deviation t-value and p-value of the slope?Try MarkusN’s r.regression.series in the addons for grass7. It allows the calculation of t and f values. IIRC, p-value can be calculated from t-value.
Moritz
r.regression.series is a module to calculate linear regression parameters between two time series. I only want to calculate parameters in one time series, how to use this model in one time serie condition?
thanks!
dehui
···
On Thu, Nov 8, 2012 at 10:55 AM, Moritz Lennert <mlennert@club.worldonline.be> wrote:
On 06/11/12 09:55, Paulo van Breugel wrote:
Hi,
I am using r.series to calculate linear regression slope, offset and
coefficient of determination. But any idea how I can get the standard
deviation t-value and p-value of the slope?Try MarkusN’s r.regression.series in the addons for grass7. It allows the calculation of t and f values. IIRC, p-value can be calculated from t-value.
Moritz
Hi Dehui,
What about my previous suggestion to use dummy variable that indicates time periods (fixed values 0, 1,2,3…N) for N time periods? I am new to time series analysis, so please let me know if the below makes sense.
So you could use your data (Y) as one time series, and the dummy time variables as the second time series (T): Y = bT + c. Resulting slope, offset and R2 between r.series and r.regression.series are close to identical. Differences are very small (a factor 10¹⁰ smaller then the actual slope or offset value) so I assume this is due to some differences in rounding (@MarkusN, do you have an idea?)
Concrete, if you have a time series of let’s say annual rainfall over 10 years. You create 10 layers using r.mapcalc, with all cells of the first layer = 0, all cells in the second layer = 1… all cells in the 10th layer = 9. Now you have two time series you can use in r.regression.series.
There is the possible problem of auto-correlation, so the t and f values calculated with r.regression.series should be considered with care. As said, I am new to time series analysis, so will need to study more how to deal with this. Not sure that is going to be easy using GRASS.
Cheers,
Paulo
···
On Thu, Nov 8, 2012 at 2:21 PM, 王德辉 <dehuiw@gig.ac.cn> wrote:
r.regression.series is a module to calculate linear regression parameters between two time series. I only want to calculate parameters in one time series, how to use this model in one time serie condition?
thanks!
dehui
-----原始邮件-----
发件人: “Paulo van Breugel” <p.vanbreugel@gmail.com>
发送时间: 2012年11月8日 星期四
收件人: “Moritz Lennert” <mlennert@club.worldonline.be>
抄送: “GRASS users email list” <grass-user@lists.osgeo.org>
主题: Re: [GRASS-user] Calculate p value for regression slope in r.seriesGood suggestion, and coincidentally I was just looking at it. Am I correct to assume that to mimic the calculation of slope etc in r.series, I would need to create a second time series of ‘dummy’ maps with values 0 for first time period, 1 for second time period, etc.? I did this in R and it gave the same slope and offset as r.series.
Just wondering, if it is implemented in r.regression.series, would it be very difficult to port to r.series? Or do the two things function different internally?
Cheers,
Paulo
On Thu, Nov 8, 2012 at 10:55 AM, Moritz Lennert <mlennert@club.worldonline.be> wrote:
On 06/11/12 09:55, Paulo van Breugel wrote:
Hi,
I am using r.series to calculate linear regression slope, offset and
coefficient of determination. But any idea how I can get the standard
deviation t-value and p-value of the slope?Try MarkusN’s r.regression.series in the addons for grass7. It allows the calculation of t and f values. IIRC, p-value can be calculated from t-value.
Moritz
Paulo van Breugel wrote:
Just wondering, if it is implemented in r.regression.series, would it be
very difficult to port to r.series? Or do the two things function different
internally?
It could probably be added to r.series quite quickly if someone was
willing to provide definitions for "t-value" and "p-value".
--
Glynn Clements <glynn@gclements.plus.com>
On Thu, Nov 8, 2012 at 6:22 PM, Glynn Clements <glynn@gclements.plus.com> wrote:
Paulo van Breugel wrote:
Just wondering, if it is implemented in r.regression.series, would it be
very difficult to port to r.series? Or do the two things function different
internally?It could probably be added to r.series quite quickly if someone was
willing to provide definitions for "t-value" and "p-value".
The t-value of a simple linear regression can be calculate with
t = sqrt(Rsq) * sqrt((n - 2) / (1 - Rsq));
Rsq = R squared or coefficient of determination (r.series method=detcoeff)
n = sample size
This formula is used by r.regression.multi and in some modules in
GRASS 5 that also do multiple regressions.
The p-value would then be the cumulative probability of observing t
according to Student's t-distribution for given n. I think somewhere
in GRASS is a function to calculate p for n = inf (normal
distribution). Maybe somewhere hidden there is also a function for
small n?
Markus M
Markus Metz wrote:
>> Just wondering, if it is implemented in r.regression.series, would it be
>> very difficult to port to r.series? Or do the two things function different
>> internally?
>
> It could probably be added to r.series quite quickly if someone was
> willing to provide definitions for "t-value" and "p-value".The t-value of a simple linear regression can be calculate with
t = sqrt(Rsq) * sqrt((n - 2) / (1 - Rsq));
Rsq = R squared or coefficient of determination (r.series method=detcoeff)
n = sample size
In which case, it can be calculated with e.g.
r.series ... output=tmp.rsq,tmp.n method=detcoeff,count
r.mapcalc 'outmap = sqrt(tmp.rsq * (tmp.n - 2) / (1 - tmp.rsq))'
I've added method=tvalue in r53742.
--
Glynn Clements <glynn@gclements.plus.com>
Thank you Paulo! I think your way is feasible. For r.series, t could be caluclate via R2 using r.mapcalc. I am also a newcomer to time series analysis. Hope r.series may add t,f, p function soon.
Best regard,
Dehui
···
On Thu, Nov 8, 2012 at 2:21 PM, 王德辉 <dehuiw@gig.ac.cn> wrote:
r.regression.series is a module to calculate linear regression parameters between two time series. I only want to calculate parameters in one time series, how to use this model in one time serie condition?
thanks!
dehui
-----原始邮件-----
发件人: “Paulo van Breugel” <p.vanbreugel@gmail.com>
发送时间: 2012年11月8日 星期四
收件人: “Moritz Lennert” <mlennert@club.worldonline.be>
抄送: “GRASS users email list” <grass-user@lists.osgeo.org>
主题: Re: [GRASS-user] Calculate p value for regression slope in r.seriesGood suggestion, and coincidentally I was just looking at it. Am I correct to assume that to mimic the calculation of slope etc in r.series, I would need to create a second time series of ‘dummy’ maps with values 0 for first time period, 1 for second time period, etc.? I did this in R and it gave the same slope and offset as r.series.
Just wondering, if it is implemented in r.regression.series, would it be very difficult to port to r.series? Or do the two things function different internally?
Cheers,
Paulo
On Thu, Nov 8, 2012 at 10:55 AM, Moritz Lennert <mlennert@club.worldonline.be> wrote:
On 06/11/12 09:55, Paulo van Breugel wrote:
Hi,
I am using r.series to calculate linear regression slope, offset and
coefficient of determination. But any idea how I can get the standard
deviation t-value and p-value of the slope?Try MarkusN’s r.regression.series in the addons for grass7. It allows the calculation of t and f values. IIRC, p-value can be calculated from t-value.
Moritz
Great, thanks
···
On Fri, Nov 9, 2012 at 12:28 PM, Glynn Clements <glynn@gclements.plus.com> wrote:
Markus Metz wrote:
Just wondering, if it is implemented in r.regression.series, would it be
very difficult to port to r.series? Or do the two things function different
internally?It could probably be added to r.series quite quickly if someone was
willing to provide definitions for “t-value” and “p-value”.The t-value of a simple linear regression can be calculate with
t = sqrt(Rsq) * sqrt((n - 2) / (1 - Rsq));
Rsq = R squared or coefficient of determination (r.series method=detcoeff)
n = sample sizeIn which case, it can be calculated with e.g.
r.series … output=tmp.rsq,tmp.n method=detcoeff,count
r.mapcalc ‘outmap = sqrt(tmp.rsq * (tmp.n - 2) / (1 - tmp.rsq))’I’ve added method=tvalue in r53742.
–
Glynn Clements <glynn@gclements.plus.com>