Re: Normal curves

shunya@xxxxxxxx wrote:
Dear James,
I do understand the measure thoretic probability ..having some
physics and a little bit
of maths background.
I plan to provide the normal graps superposed on the observed
graphs for " data quality
assurance".
This is only a sound approach if you already know that the underlying distributions are normal.

Let me see if I can give you some intutition here. Suppose that there are errors and they are normally distributed, and there is the true data which is normally distributed. The observations constitute a sum of two normally distributed variables, true data plus error, which means that the observations will appear normal even if there are errors. If this is the case, you'll accept bad data as though it were good.

On the other hand, if the generating process creates skewed true data, then your data will not be normally distributed if it is accurate. If this is the case and you use normality as your criterion for data accuracy, you'll reject good data as though it were bad.

I have a data entry program with a lot of whistles . Ringing bells
are missing.
Why would the data % not be normally distributed ? if the values are
discretised then they will only give small peaks/dumps spread out on
the normal curve.


On a lot of the Questions on the same curve the superposition is just
excellent ..
Contacting a statistical consultant is the most remote possibility.
I used SPSS on a big huge iron box long back.
Pointers to some reading material will help me.
You got it. If you are near a large library, look for a book called "handbook of statistical distributions." You will probably find more than one book by this title. Any book by this title is a good starting point for what you are describing.

Usenet is /was for the inquisitve..
thanks.
Mahesh Naik

James_W wrote:

shunya@xxxxxxxx wrote:

Hello reader,
I am looking for experience derived from actual field surveys(the
demographic variety) about the normality of values of question
responses. Question like age , number of children
etc etc.
I had a dataset of 8000 (approx.) individuals consisting of 4
subpoulations. On plotting
the observed poulation % and the expected normal % , i saw that there
were more peaks and dumps then expected.[ may be the investigators were
not 'not lazy'].
Which stastical values do you use to describe the data in such
situations, other than population %s.
Mahesh Naik
ps: I donot understand probability and cant understand things like
confidence intervals.


If you don't understand probability, then you really need to understand
that first.

The variables you've mentioned will almost certainly not be normally
distributed, but the numbers used to describe these populations vary
depending on the distributions, which will be different for different
variables.

So here's the next question: What are you planning on doing with this
data? If you are preparing a presentation for a non-technical audience,
then some % graphs will probably be ok. If you want to do some heavy
number crunching, you probably need a statistical consultant, not usenet.

Regards,
James

--
Politics, philosophy and economics, from a libertarian perspective.
http://n-k-1.blogspot.com




--
Politics, philosophy and economics, from a libertarian perspective.
http://n-k-1.blogspot.com
.