In article <[EMAIL PROTECTED]>,
Stan Brown  <[EMAIL PROTECTED]> wrote:
>les <[EMAIL PROTECTED]> wrote in sci.stat.edu:
>>i have 327 data points and i plotted the histogram which
>>looks like poisson distribution. I want to test the hypothesis that
>>this distribution is significantly different from uniform and want to get 
>>get some confidence level (say 0.95 ). I was thinking of chi square but 
>>does'nt chi square assume normal distributed data? 

>Not necessarily. Chi-square compared the observed number of data 
>points in each "bucket" with the expected number in each "bucket". 
>You would set up your "buckets" and then calculate how many data 
>points out of 327 would fall into each "bucket" if the distribution 
>is what you want to test, such as Poison.

>What I don't know is how you decide how many "buckets" to use. 
>Intuition tells me that the number of "buckets" may influence 
>whether you get a significant difference or not, since in the 
>limiting case of n=1 you obviously get no difference at all.

The chi-squared test of goodness of fit is generally a very
poor test, with low resolving power.  This is because it
completely ignores the ordering of the data points.

The test statistic only asymptotically has a chi-squared
distribution, and this is only if one tests from a fixed
distribution, which means that no parameters are
estimated.  If the parameters are estimated from the
"bucket" proportions ONLY, the asymptotic distribution
is still chi-squared with a reduced number of degrees of
freedom.  Other tests are more powerful, especially 
parametric tests.
-- 
This address is for information only.  I do not claim that these views
are those of the Statistics Department or of Purdue University.
Herman Rubin, Deptartment of Statistics, Purdue University
[EMAIL PROTECTED]         Phone: (765)494-6054   FAX: (765)494-0558
.
.
=================================================================
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