| |
 |
|
|
Science Forum Index » Statistics - Math Forum » asymmetric gaussian
Page 1 of 1
|
| Author |
Message |
| mauro |
 Posted: Wed Apr 02, 2008 4:43 pm |
|
|
|
Guest
|
Hi all,
I got a vector of data approximately gaussian distributed.
I observed that the data doesn't form not a perfect gaussian, in fact
once represented, the shape is a kind of asymmetric gaussian, with the
left part less sloped then the right one.
I'm searching a mathematical model to describe it better then a simple
gaussian.
Using something like left and right variance. Does it exist something
like that?
Do you have any idea?
cheers,
Mauro |
|
|
| Back to top |
|
| John Uebersax |
| Posted: Wed Apr 02, 2008 7:26 pm |
|
|
|
Guest
|
One common method is to model it as a mixture of two gaussians. To
see and example, check
this page:
http://ourworld.compuserve.com/homepages/jsuebersax/skewed.htm
and scroll down to Figure 2.
HTH
John Uebersax PhD
On Apr 3, 4:43 am, mauro <mauro.austra...@gmail.com> wrote:
Quote: Hi all,
I got a vector of data approximately gaussian distributed.
I observed that the data doesn't form not a perfect gaussian, in fact
once represented, the shape is a kind of asymmetric gaussian, with the
left part less sloped then the right one.
I'm searching a mathematical model to describe it better then a simple
gaussian.
Using something like left and right variance. Does it exist something
like that?
Do you have any idea?
cheers,
Mauro |
|
|
| Back to top |
|
| Ray Koopman |
| Posted: Wed Apr 02, 2008 9:42 pm |
|
|
|
Guest
|
On Apr 2, 7:43 pm, mauro <mauro.austra...@gmail.com> wrote:
Quote: Hi all,
I got a vector of data approximately gaussian distributed.
I observed that the data doesn't form not a perfect gaussian, in fact
once represented, the shape is a kind of asymmetric gaussian, with the
left part less sloped then the right one.
I'm searching a mathematical model to describe it better then a simple
gaussian.
Using something like left and right variance. Does it exist something
like that?
Do you have any idea?
cheers,
Mauro
Another way is by probabilistic selection. Let x and y have a joint
normal distribution, and consider the distribution of y-scores for
people whose x-score is greater than some arbitrary cutoff. |
|
|
| Back to top |
|
| David Jones |
| Posted: Thu Apr 03, 2008 5:15 am |
|
|
|
Guest
|
mauro wrote:
Quote: Hi all,
I got a vector of data approximately gaussian distributed.
I observed that the data doesn't form not a perfect gaussian, in fact
once represented, the shape is a kind of asymmetric gaussian, with the
left part less sloped then the right one.
I'm searching a mathematical model to describe it better then a simple
gaussian.
Using something like left and right variance. Does it exist something
like that?
Do you have any idea?
cheers,
Mauro
There are all sorts of posibilities, including
(i) the standard families of distributions (Pearson, Johnson, etc.) should contain possibilities;
(ii) a simple power transformation might give something useful.
However, thinking about the context of your dataset is probably the place to start as there may be some particular relevance of certain transformations (eg squaring) in you particular context ... also relevant are any lower- or upper-bounds that the values must respect.
David Jones |
|
|
| Back to top |
|
| mauro |
| Posted: Tue Apr 08, 2008 11:53 pm |
|
|
|
Guest
|
On Apr 3, 12:43 pm, mauro <mauro.austra...@gmail.com> wrote:
Quote: Do you have any idea?
[...]
thanks everybody for the answers (and sorry for the late reply,
unfortunately I've have had a little accident..)
The idea of the sum of two normal curve is the one I needed!
Anyway (David asked me) about the context of the dataset: the values
are from a part of a CT scan of the head (a nerve).
cheers,
Mauro |
|
|
| Back to top |
|
| aruzinsky |
| Posted: Wed Apr 09, 2008 5:55 am |
|
|
|
Guest
|
On Apr 2, 8:43 pm, mauro <mauro.austra...@gmail.com> wrote:
Quote: Hi all,
I got a vector of data approximately gaussian distributed.
I observed that the data doesn't form not a perfect gaussian, in fact
once represented, the shape is a kind of asymmetric gaussian, with the
left part less sloped then the right one.
I'm searching a mathematical model to describe it better then a simple
gaussian.
Using something like left and right variance. Does it exist something
like that?
Do you have any idea?
cheers,
Mauro
There is the "Epsilon-Skew-Exponential Power Distribution" or, more
specifically, "Epsilon-Skew-Normal" . About six months ago, I
downloaded a free PDF by Hassan Elsalloukh, but it is apparently no
longer free or available.
http://ualr.edu/hxelsalloukh/publication.html |
|
|
| Back to top |
|
| |
|
Page 1 of 1
All times are GMT - 5 Hours
The time now is Thu Jul 24, 2008 4:59 pm
|
|
