Re: Gaussian distribution and likelihood question
- From: David Winsemius <doe_snot@xxxxxxxxxxx>
- Date: Mon, 29 Jan 2007 18:28:53 -0600
"zl2k" <kdsfinger@xxxxxxxxx> wrote in news:1170093960.945613.276350
@p10g2000cwp.googlegroups.com:
hi,
My question is: is there any method to measure how a data set fits a
Gaussian model? Suppose I have one dataset A which has N elements and
another dataset B which has M elements. From each of the dataset I can
estimate one Gaussian model, thus I have (mu1, sigma1) for A and (mu2,
sigma2) for B. How can I tell which dataset fits the Gaussian model
better? (skewness and kurtosis does not tell about fitness)
What I am thinking is from the Gaussian model I can generate the same
amount of elements and those elements should fit the model best. Then
I can compare the log-likelihood of the real dateset to the generated
dataset. But, to produce a psudo dataset from a Gaussian model is not
efficient. My next question is: is there a 1-step formula to get the
log-likelihood of the "theorically perfect Gaussian dataset" with N
elements? Basically, find the most possible log-likelihood of N
elements from Gaussian model(mu, sigma).
Suggest a quantile-quantile plot.
--
David Winsemius
.
- Follow-Ups:
- References:
- Prev by Date: Re: Improper distribution
- Next by Date: Re: Multiple exponential distributed arrival rates
- Previous by thread: Gaussian distribution and likelihood question
- Next by thread: Re: Gaussian distribution and likelihood question
- Index(es):
Relevant Pages
|
|
