On 2/9/2010 1:43 AM, Eli wrote:
> 1)
> I have a vector of values 'Noise' that should have a Rayleigh distribution.
> I would like to preform a test to determine the how probable it is that
> this vector was in fact sampled from a Rayleigh distribution. the vector
> is 50176 elements in length.
> What is the best way to do this?
The KSTEST function in the Statistics Toolbox might be one way to do that. But the KS test assumes that the distribution you're testing against is fully known in advance. The pvalue will be incorrect if you estimate the parameters of a Rayleigh distribution from your data, and then do a KS test on those same data, using the estimated distribution as your null hypothesis. Such a test would typically be conservative, meaning it would tend to reject less than it should.
> 2)
> I have another vector 'Gnoise' that should have a gaussian distribution,
> but not centred at zero. I would like to preform a test to determine the
> how probable it is that this vector was in fact sampled from a Rayleigh
> distribution. This vector is 10201 elements in length.
> What is the best way to do this?
I assume you mean, "was in fact sampled from a Gaussian distribution". The LILLIETEST function in the Statistics Toolbox is one possibility. This test has a builtin compensation for the problem mentioned above, so you use to test against "unspecified normality".
Hope this helps.
