Thank for your reply!

"Rich Ulrich" <[EMAIL PROTECTED]> wrote in message
news:[EMAIL PROTECTED]
> On Mon, 01 Dec 2003 02:33:13 GMT, "jackson marshmallow"
> <[EMAIL PROTECTED]> wrote:
>
> > Hello everyone,
> >
> > I hope I can get simple answers to these questions... I need to solve a
> > couple of practical problems and I'm new to statistics...
> >
> > 1) Two samples of are given and I need to compare their means and
variances.
>
> I did not notice this at first reading -- you say that you
> want to compare the variances, too.
>

I'm not sure yet, but I may need to look into that...

> Variance differences make it difficult to do tests of means,
> and more, it can make it difficult to interpret them.  The
> group with REALLY big variance will have the extreme
> scores in both directions:  Otherwise, you are talking
> about something that I would consider to be confounded
> with *scaling* questions.    Are you absolutely sure that
> you are interested in the difference of the 'arithmetic
> mean'? and not the 'central tendency', or the superiority
> at one end or the other?
>

The thing is, there's more than one problem here... In one case, I may need
to compare means, in another variances...

>
> So, look at your plots.  Similar?  Shifted?  Expanded?
> You could do several varieties of tests and see if they
> all come out the same, as a cheap version of 'testing for
> normality.'   That's the easiest step, when you don't want
> to commit to anything about the numbers.
>

Ultimately the basic statistical calculations will have to be automated for
large amount of experiments, and I will not be able to build / examine plots
for most of them. Of course, it won't hurt to do a sanity check every now
and again...

> Oh, the basic, simple, usual nonparametric tests that use
> ranks  have assumptions that are almost as harsh as those
> of the t-test, so you can't really opt out of all thinking merely
> by employing that rank transform.
>

Yes, so I understand.  That would be too easy :)

>
> > The distribution of the population is unknown. Can I use the F-test and
the
> > t-test? Is it necessary that the sample _means_ have a Gaussian
> > distribution? Is it sufficient? Maybe I misunderstand something here...
>
> If there are no outliers, then the t-test is pretty robust
> for the means.  A test on ranks is not exactly a test on means.
>
> > 2) I need to calculate the significance of correlation between two
> > sequences. I would actually prefer to use randomization, but the
sequences
> > may be too short. Another option is to perform linear regression and
> > calculate the significance of the slope using a t-test (?). When is it
> > valid?
>
> As Dave says -- if you have a sequence where the word
> 'sequence'  is meaningful, then you have to take that into

"sequences of numbers" is all I meant; maybe I'm misusing the term

> account; simple zero-order  correlation (or regression) gives
> bogus values and tests when there is dependency.
>
>
> --
> Rich Ulrich, [EMAIL PROTECTED]
> http://www.pitt.edu/~wpilib/index.html
> "Taxes are the price we pay for civilization."


.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to