Bootstrap method

[Petcu Daniel]

Dear Members,

Sorry for a stupid question !
Is the bootstrap method, as described in Lenhoff MW et al. "Bootstrap prediction and confidence bands: a superior statistical method for analysis of gait data" Gait Posture. 1999 Mar;9(1):10-7., implemented in a software which can be used to analyze some data ?
Or, there is any software which have a suitable bootstrap method implemented for the analysis of the force platform data ?

Thank you !
Respectfully,
Daniel

[Dan Robbins]

Re: Bootstrap method

Hi Daniel,

Certainly not a stupid question - its always worth asking for help when you're trying to figure something out!

Matlab has bootstrap and bootstrap confidence interval functions that you could try, I am not familiar with the article or your specific goal, so I can't comment on the exact way the code should be constructed. This is at least something you could look in to though.

Hope it helps
Dan

[Petcu Daniel]

Re: Bootstrap method

Hi Dan,

First, thanks a lot for the encouraging reply!
My goal is to compare the effect of a conservative intervention [ex. footwear] on the evolution of the free moment in the entire or partial intervals of the stance phase of the gait cycle. The first reference on this subject I found in the article: http://www.ncbi.nlm.nih.gov/pubmed/20870512 . As the bootstrap method is seen as a "standard method for evaluating gait data curves" I've decided to learn about it. I've seen the method implemented also in IBM SPSS or R and even as Excell add-in . But the mentioned article, http://www.ncbi.nlm.nih.gov/pubmed/10575065 , start from a finite Fourier representation of the kinematic / kinetic curves which I don't understand if it is applyed in the already existent solution from the mentioned software.

Any help will be apreciated !

Respectfully,
Daniel

[Dan Robbins]

Re: Bootstrap method

Hi Daniel,

The type of study you are describing has been discussed in a few threads on here - have you searched the forum? One thread particularly comes to mind http://biomch-l.isbweb.org/threads/2...=wavelet+anova which could be of use. I would suggest there are a variety of approaches you could adopt to get the results you are looking for. It also depends on the resources you have available of course, so perhaps have a look at the different approaches and see which is appealing and practical.

Hope that helps
Dan

[Philippe C. Dixon]

Re: Bootstrap method

Dear Daniel,

I have recently used the Bootstrap method described by Lenhoff et al. to analyse differences between conditions over the entire gait cycle (Dixon et al. 2013. Spatio-temporal parameters and lower-limb kinematics of turning gait in typically developing children. Gait Posture. 38(4):870-5). My code relies mainly on the MatLab function "datasample" to resample data with replacement. I would be happy to share it with you if you like.

Best Wishes,

Phil

[Petcu Daniel]

Re: Bootstrap method

Hello Dan,

Thank you again for your useful replay. From the indicated thread, it was interesting for me the article: http://www.ncbi.nlm.nih.gov/pubmed/22178005 . This is using the functional analysis of variance [FAOV]. Taking into account that I can't benefit from the help of a statistician I'll see which one from the 2 methods, FFOV and bootstrap, could be easily learned and implemented in my case. As the functional data analysis has the benefit of the website http://www.psych.mcgill.ca/misc/fda/index.html as source of information I'm tempted to start with this method.

Any other advice will be apreciated !

Respectfuly,
Daniel

[Petcu Daniel]

Re: Bootstrap method

Dear Daniel,

I have recently used the Bootstrap method described by Lenhoff et al. to analyse differences between conditions over the entire gait cycle (Dixon et al. 2013. Spatio-temporal parameters and lower-limb kinematics of turning gait in typically developing children. Gait Posture. 38(4):870-5). My code relies mainly on the MatLab function "datasample" to resample data with replacement. I would be happy to share it with you if you like.

Best Wishes,

Phil

Dear Philippe,
Thank you for your answer!
It will be great for me to benefit from your help !
How can I proceed to get it ?Respectfully,
Daniel

[Dan Robbins]

Re: Bootstrap method

Dear Phil,

I am also interested in your approach here, did you use matlab to process all of the data or did you preprocess the data with different software, for example visual3D?

I am particularly interested if you put raw kinematic data directly in to matlab or if you normalised/identified the start/end of the gait cycles before processing in the data matlab.

Regards

Dan

[Petcu Daniel]

Re: Bootstrap method

Dear Phil,
Thanks for the reference! Very interesting article ! The first thing I've looked for, it was the force platform because the article I've mentioned ( http://www.ncbi.nlm.nih.gov/pubmed/20870512 ) is analysing pivoting tests for normal and injuried knees using, in my understanding, a point-by-point Gaussian-based confidence intervals methodology. A suplementary reason for trying to find a solution with boostrap method !
Respectfuly,
Daniel

[Patrick Salvia]

Re: Bootstrap method

Dear Philippe and All,

In biomechanics, not only engineers are interested in curves of patterns analysis. The medical doctors, surgeons or physiotherapists or students understand the principle of the method but they don't have all the mathemetical background to implement this in Matlab without an example of the code. They look in Matlab Central or in Biomch Software to find a way. And they lose a lot of time to really give an answer to the reaserch question.
I tried to contact Lenhoff to ask if they have developed a Matlab or any other language code for the application of this method, but without any response . I also contacted A. Duhamel for his statistical tools for clinical gait analysis.Duhamel A, et al Gait Posture . 2004 : 204-12 without any response .
I have also found this toolbox http://www.csp.curtin.edu.au/downloa...p_toolbox.html for bootstrapping.

I do not know if it's enough to really test whether bootstrapping approach gives more confidence than 2 or 3 standard deviations on the motion curves classicaly used in medical and biomechanical papers for Gaussian data.

Thank you if you can share something,




Patrick Salvia

Laboratory of Anatomy, Biomechanics and Organogenesis,
Université Libre de Bruxelles

Dear Daniel,

I have recently used the Bootstrap method described by Lenhoff et al. to analyse differences between conditions over the entire gait cycle (Dixon et al. 2013. Spatio-temporal parameters and lower-limb kinematics of turning gait in typically developing children. Gait Posture. 38(4):870-5). My code relies mainly on the MatLab function "datasample" to resample data with replacement. I would be happy to share it with you if you like.

Best Wishes,

Phil

[Philippe C. Dixon]

Re: Bootstrap method

Dear Philippe and All,

In biomechanics, not only engineers are interested in curves of patterns analysis. The medical doctors, surgeons or physiotherapists or students understand the principle of the method but they don't have all the mathemetical background to implement this in Matlab without an example of the code. They look in Matlab Central or in Biomch Software to find a way. And they lose a lot of time to really give an answer to the reaserch question.
I tried to contact Lenhoff to ask if they have developed a Matlab or any other language code for the application of this method, but without any response . I also contacted A. Duhamel for his statistical tools for clinical gait analysis.Duhamel A, et al Gait Posture . 2004 : 204-12 without any response .
I have also found this toolbox http://www.csp.curtin.edu.au/downloa...p_toolbox.html for bootstrapping.

I do not know if it's enough to really test whether bootstrapping approach gives more confidence than 2 or 3 standard deviations on the motion curves classicaly used in medical and biomechanical papers for Gaussian data.

Thank you if you can share something,




Patrick Salvia

Laboratory of Anatomy, Biomechanics and Organogenesis,
Université Libre de Bruxelles

Dear Daniel, Dan, Patrick et al.,

I have uploaded an m-file that computes Bootstrapped confidence intervals to the MatLab file exchange . Please find the link below and do not hesitate to contact me with any further questions or if you find bugs in the function. The function takes partitioned (to gait cycle for example) and normalized data as input.

http://www.mathworks.com/matlabcentral/fileexchange/44916-bootstrap-gait-curves

Best Wishes,

Phil

[Petcu Daniel]

Re: Bootstrap method

Dear Philippe,

Thank you very much for help !
I'm not a statistician but I have a question regarding the Lenhoff's article : is the sample variance-covariance's formula corect ? I'm thinking that the division is made by "n" as in the case of a population and not by "n-1" as in the case of a sample. Or in the article is stated that "the sample variance-covariance matrix ....is an estimator of the population variance-covariance"

Sincerely,
Daniel

[Patrick Salvia]

Re: Bootstrap method

Hi Daniel, Phil and Other,

Thanks also Phil for the m file.

I have not the time to check the formula , but have a look of this paper
Gait Analysis and the bootstrap by Richard Olshen and also David Sutherland in the Annals of Statistics 1989, 17 :4:1419-1440

Best Regards

Patrick

Dear Philippe,

Thank you very much for help !
I'm not a statistician but I have a question regarding the Lenhoff's article : is the sample variance-covariance's formula corect ? I'm thinking that the division is made by "n" as in the case of a population and not by "n-1" as in the case of a sample. Or in the article is stated that "the sample variance-covariance matrix ....is an estimator of the population variance-covariance"

Sincerely,
Daniel

[Petcu Daniel]

Re: Bootstrap method

Hi Patrick,

To be onest is not an easy task for me to understand very quickly Olshen et. all demonstration. Knowing that in Excel it is a problem with covariance formula (because of using "n" instead of "n-1") and also from this source (for example): http://en.wikipedia.org/wiki/Sample_mean_and_covariance , I suppose it could be a problem because the Lenhoff article's is speaking about "the sample variance-covariance matrix ...., defined as 1/n*(...)" which "is an estimator (biased) of the population variance-covariance Ew"
Hope you'll find some time to check the formula...
Sincerely,
Daniel

[Govindaraju Kalimuthu]

Re: Bootstrap method

Dear Mr Daniel,
I have gone through the paper as you mentioned below. Did you find the solution for Bootstrap method. Which software useful to get confidence and prediction band for gait analysis.

IS it MATLAB or any other software ?

Thank you for your time

Govind

Dear Members,

Sorry for a stupid question !
Is the bootstrap method, as described in Lenhoff MW et al. "Bootstrap prediction and confidence bands: a superior statistical method for analysis of gait data" Gait Posture. 1999 Mar;9(1):10-7., implemented in a software which can be used to analyze some data ?
Or, there is any software which have a suitable bootstrap method implemented for the analysis of the force platform data ?

Thank you !
Respectfully,
Daniel