Which equation does Matlab use to fit a Normal distribution?
2 views (last 30 days)
Show older comments
Andreina Urquiola
on 11 Feb 2021
Commented: Andreina Urquiola
on 15 Feb 2021
Which equation does Matlab use to fit a Normal distribution?, especially for fitdist(x,'Normal') function. I would like to know if the fitting is given by:
or by another expression.
2 Comments
Accepted Answer
Matt J
on 11 Feb 2021
Edited: Matt J
on 11 Feb 2021
To reverse engineer, we can fit the following data
x=randn(1e5,1)*2+5;
which is definitely distributed according to the first density parametrization that you posted (what I would consider the classical parametrization) with mu=5 and sigma=2. Since the routine gives us virtually the same sigma and mu as what we simulated, there is a good chance that the parametric model is the classical one:
pd = fitdist(x,'Normal')
2 Comments
Matt J
on 12 Feb 2021
You're welcome, but please Accept-click an answer that you deem best solves your question.
More Answers (1)
John D'Errico
on 11 Feb 2021
I think you misunderstand, as there are several things happening here.
A normal distribution fit, that is, finding the parameters of a Normal PDF assumes the distribution has the property that the integral of that function is 1. A PDF has that property, and this is implicit in tools like fitdist.
However, IF you use some other software to perform a nonlinear regression fit, to a model as you show, thiis is NOT a normal distribution. It fails the property that the integral of that function is 1. In fact, the integral from -inf to inf is unbounded, if y0 is ANY number other than zero.
So it is NOT a normal distribution. You may call it a Gaussian fit, but the only thing it has in common with a normal distribution is it looks like a normal, and it is based on the same equation, though perhaps with some additional parameters. It is NOT a normal distribution.
See Also
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!