Hi Radford, thankyou kindly for your response. I took a statistics course long time ago (around 5 years) so i have been using the terminology incorrectly. Yes, there is an upper bound to the data and i should have said binomial. However, for the null hypothesis i am not sure if p=0.5 is appropriate since i want to actually test for hypothesis that there is no dependence on the input variable. i am measuring some discrete characteristic of a some object as a function of the discrete length (i.e. 1 unit, 2units etc) and the when i plot this measure (which is length normalized measure) i see that there is some correlation i.e. it looks like a binomial distribution with p = 0.1 . I had expected to see that this measure should not vary with length . Does'nt this mean that I should calculate some measure of nonuniformity? (i.e. compare it Uniform(0,Length) ) ? thanks
[EMAIL PROTECTED] (Radford Neal) wrote in message news:<[EMAIL PROTECTED]>... > In article <[EMAIL PROTECTED]>, > les <[EMAIL PROTECTED]> wrote: > > >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? > > "chi square" is the name of a distribution, which is often used as the > name of a test that uses that distribution. Unfortunately, there are > several such tests. One of these is for the variance of a normal > distribution, which doesn't apply to your situation. Another "chi square" > test, however, is for the fit of discrete data to a known distribution. > This would be appropriate for you. > > At least it would be appropriate if the hypothesis you describe is > actually appropriate. I wonder about that, though. You say the data > "looks like a poisson distribution". A poisson distribution has no > upper limit. You say you want to test the null hypothesis that the > data comes from a uniform distribution. Uniform over what set of > values? There has to be an upper limit to a uniform distribution. > > If there is some known upper limit, then perhaps you should really be > thinking of the data as "looking" like it comes from a binomial > distribution. At that point, one might wonder whether the null > hypothesis you should really be testing is that the data is binomial > with p=1/2. (I'm assuming your data consists of 327 non-negative > integers, since you mention a Poisson distribution.) > > In general, it's dangerous to ask advice of this nature without saying > anything about your real problem. You may well get advice that sounds > authoratative (and maybe is, in a narrow sense), which then encourages > you to think you're doing the right thing, when actually you're totally > confused. > > By the way, the usual way of expressing the result of a hypothesis > test is a "p-value". By "confidence level", I can only assume you > mean one minus this p-value. However, you shouldn't say that you > "want to get some confidence level (say 0.95)". You do the test, and > you get some p-value. Your wants don't come into it. It's not like > finding a confidence interval, where you can choose the confidence > level. > > Radford Neal > > ---------------------------------------------------------------------------- > Radford M. Neal [EMAIL PROTECTED] > Dept. of Statistics and Dept. of Computer Science [EMAIL PROTECTED] > University of Toronto http://www.cs.utoronto.ca/~radford > ---------------------------------------------------------------------------- . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
