Can we apply non-linear smoothing to a linear looking like data ?

[paawansharmas]

My doubt is that whether we can apply non-linear smoothing to a almost linear data ( without one or 2 discontinuity)

I have attached the pic in which the red data is the smoothed one. Blue is the original one.
I multiplied each point with an increasing like 1, 1.1, 1.2, 1.3, 1.4...so on

My question is : is this valid way to remove discontinuity or to make a linear correlation stronger ??

thanks

 

[Amir Livne]

It really sounds like a question in the science of statistics.

For some series it'll make the linear correlation stronger, for some weaker. It probably depends on the effect you want, or the features you want to find in your data.

Statistics can also help you choose the correct coefficients for whatever kind of dampening.

So I advise you to ask this again in (or move the thread to) the statistics forum.

 

[paawansharmas]

Thanks Amir.
But i do not know how to move it to other forum.
and if I write same thread in another forum , I will get warnings for multiple posting.

please help.

Thanks
Paawan

 

[Stephen Tashi]

My doubt is that whether we can apply non-linear smoothing to a almost linear data ( without one or 2 discontinuity)

To move a thread, you can use the "report" feature (even though the directions for report sound like it is only to be used to report naughty things) and "report" that you would like your thread moved.

You have not defined a mathematical problem. It isn't clear what you are trying to accomplish.

To take a silly point of view, you can do anything you want with the data. You could erase each value and write in a different number that pleases you! However, your question suggests you think only certain procedures will accomplish your goal. But what is the goal?

 

[ssd]

The answer to your question is 'yes'. Your curve looks like having a quadratic component. Take your model as Y= a+ bx + cx^2. Minimize sum[(y-Y)^2] with respect to a,b,c, where y= observed values.