home *** CD-ROM | disk | FTP | other *** search
- Comments: Gated by NETNEWS@AUVM.AMERICAN.EDU
- Path: sparky!uunet!uvaarpa!darwin.sura.net!gatech!paladin.american.edu!auvm!APG-EMH7.ARMY.MIL!LEE_RB
- Message-ID: <STAT-L%93012616452854@VM1.MCGILL.CA>
- Newsgroups: bit.listserv.stat-l
- Date: Tue, 26 Jan 1993 16:35:00 EST
- Sender: STATISTICAL CONSULTING <STAT-L@MCGILL1.BITNET>
- From: "MARS::LEE_RB" <LEE_RB%MARS.decnet@APG-EMH7.ARMY.MIL>
- Subject: Chi-square test of a Binomial Distribution
- Lines: 17
-
- Here is my dilema: I have n=4 animals with t=100 trials each. Each animal
- selects either a familiar or novel object on each trial. Each animal's
- probability of selecting the familiar object (or the novel object) follows a
- binomial distribution. The number of choices in which the animal selects
- the familiar object is X. Therefore, each animal's probability of selecting
- the familiar object is p=X/t with variance tp(1-p).
- Now I am reading a paper that states that the mean of the animals
- probabilities is P with expected variance tP(1-P) and the observed variance is
- calculated using each animal's p according to the equation for variance for
- a normal distribution. It is further stated that the ratio of the observed to
- the expected multiplied by (n-1) will yield a Chi-square statistic.
- Now I understand that if I get a large Chi-square, I will reject the
- hypothesis that the animals as a group do not fit a binomial distribution.
- But what does it tell me when I get and extremely low Chi-square and can I
- place a probability value on getting such a value? This paper states that an
- extremely low Chi-square would also be indicative of not fitting a binomial
- distribution. Is this really true?
-