Prev: multidimensional data in svmclassify
Next: Improving precision in fminunc
From: Samoline1 Linke on 7 Oct 2009 07:11
Hello,

I have to do some statistical analysis with my data but my data is not normally distribtued.

Is there any way to make it close to normal distribution?

I have statistical toolbox but can't see any tool to make it normally distributed.

I shall be thankful for your help.

Regards,
Sam
From: Jos on 7 Oct 2009 07:18
"Samoline1 Linke" <fixed-term.jehandad.khan(a)de.bosch.com> wrote in message <haht05$rq2$1(a)fred.mathworks.com>...
> Hello,
>
> I have to do some statistical analysis with my data but my data is not normally distribtued.
>
> Is there any way to make it close to normal distribution?
>
> I have statistical toolbox but can't see any tool to make it normally distributed.
>
> I shall be thankful for your help.
>
> Regards,
> Sam

What kind of statistical analyses are you talking about? If applicable, did you consider non-parametric approaches?

Jos
From: Samoline1 Linke on 7 Oct 2009 11:19
Thanks for the reply.

I have to do firstly pearson correlation, then regression and Anova. I have not tried non-parametric tests as my first preference are the parametric ones. Does Matlab convert the data to a normal dist? or do we have to apply non-para tests now?

There is something like dist. fitting tool in statistical toolbox but it only gives the value of mean and std. deviation and does not convert the data to normal dist.

What can i do in this case? what do you suggest please.

P.S. My data is based on transient signals (e,g, temperature, humidity etc.)

"Jos " <#10584(a)fileexchange.com> wrote in message <hahtda$r1g$1(a)fred.mathworks.com>...
> "Samoline1 Linke" <fixed-term.jehandad.khan(a)de.bosch.com> wrote in message <haht05$rq2$1(a)fred.mathworks.com>...
> > Hello,
> >
> > I have to do some statistical analysis with my data but my data is not normally distribtued.
> >
> > Is there any way to make it close to normal distribution?
> >
> > I have statistical toolbox but can't see any tool to make it normally distributed.
> >
> > I shall be thankful for your help.
> >
> > Regards,
> > Sam
>
> What kind of statistical analyses are you talking about? If applicable, did you consider non-parametric approaches?
>
> Jos
From: the cyclist on 7 Oct 2009 11:49
"Samoline1 Linke" <fixed-term.jehandad.khan(a)de.bosch.com> wrote in message <haht05$rq2$1(a)fred.mathworks.com>...
> Hello,
>
> I have to do some statistical analysis with my data but my data is not normally distribtued.
>
> Is there any way to make it close to normal distribution?
>
> I have statistical toolbox but can't see any tool to make it normally distributed.
>
> I shall be thankful for your help.
>
> Regards,
> Sam

It is not at all clear to me what you mean by "making it close to normal". The data are what they are. You cannot make them normally distributed if they are not.

It is true that a lot of techniques assume normality, but many give good approximations for data that are not normal. Maybe you could tell us a bit more about what your data look like, and what you are trying to accomplish?
From: Samoline1 Linke on 7 Oct 2009 12:26
ok thanks for taking interest.

my data is very large, u can say minimum of 10-15K entries. These are signals which change with each passing second. I have to apply pearson correlation, linear regression etc.
All these tests require data to be from Guassian/normal distribution. I have read in some books that if ur data is not normally dist. then don't expect to get accurate results from above mentioned methods. I actually want to be more accurate in my findings. I have applied normality tests and have found my data is not normally distributed.

I have also read that applying some techniques e.g. taking log, square root, or inverse etc. of the data and removing outliers makes it close to normal distribution which means we can apply later on the pearson correlation, linear regression etc.

I have tried above mentioned techniques but some of the variables are now close to normal dist. but not all. This means I might have to apply non parametric tests or have to search further how to make my data close to normal distribution.

Till now I have not looked at non parametric tests in statistical toolbox of matlab and i m not sure if matlab has any...Just saw spearman (non-parametric) as an alternative of pearson.

In short, you can say I wanna apply below mentioned methods to my data:

CORRELATION
REGRESSION
ANOVA
ANCOVA

and my data is not normally distributed but the data size is quite big. Shall I consider applying above mentioned methods or do I have to apply non parametric tests to find the correct relationship.

Kindly ask me again if u r still skeptic abotu what I have told... I would love to discuss further.





"the cyclist" <thecyclist+mathworks_newsgroup(a)gmail.com> wrote in message <haid9d$mcm$1(a)fred.mathworks.com>...
> "Samoline1 Linke" <fixed-term.jehandad.khan(a)de.bosch.com> wrote in message <haht05$rq2$1(a)fred.mathworks.com>...
> > Hello,
> >
> > I have to do some statistical analysis with my data but my data is not normally distribtued.
> >
> > Is there any way to make it close to normal distribution?
> >
> > I have statistical toolbox but can't see any tool to make it normally distributed.
> >
> > I shall be thankful for your help.
> >
> > Regards,
> > Sam
>
> It is not at all clear to me what you mean by "making it close to normal". The data are what they are. You cannot make them normally distributed if they are not.
>
> It is true that a lot of techniques assume normality, but many give good approximations for data that are not normal. Maybe you could tell us a bit more about what your data look like, and what you are trying to accomplish?
 |  Next  |  Last
Pages: 1 2
Prev: multidimensional data in svmclassify
Next: Improving precision in fminunc