home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!noc.near.net!hri.com!spool.mu.edu!darwin.sura.net!paladin.american.edu!auvm!VTVM1.BITNET!WLTPIRIE
- From: WLTPIRIE@VTVM1.BITNET (Walt Pirie)
- Newsgroups: bit.listserv.stat-l
- Subject: Chi-Squares
- Message-ID: <STAT-L%92110817120495@VM1.MCGILL.CA>
- Date: 8 Nov 92 22:06:37 GMT
- Sender: "STATISTICAL CONSULTING" <STAT-L@MCGILL1.BITNET>
- Lines: 70
- Comments: Gated by NETNEWS@AUVM.AMERICAN.EDU
-
- A note of caution on the comment below:
-
- Subtracting chi-squares to get a resultant chi-square difference
- requires more than that the originals are for different models
- on the same data set. For example, if the original two are
- "nested" models in a log-linear analysis, and the chi-squares
- are likelihood ratio chi-squares, then it will work. If the
- models are not nested, then it won't, and even if they are, if
- Pearson chi-squares are used, it won't work even then.
- What is being invoked in that argument is that the sum of two
- chi-squares is again chi-square, but that is only valid if the
- two being summed are independent.
- The original inquiry did not give any information about models
- or data sets, so there's no way to know. But if the chi-squares
- were for different data sets, then they would be independent.
- My guess is that was the case. Would their difference then be
- chi-square? That would be true if it were independent of the
- one being subtracted, but I suspect that would not be true.
- I still feel the most important point on the original inquiry
- is one raised by someone a few days ago, I don't remember who.
- These were approximate chi-squares for frequency data, and as such
- are approximations for which exact p-values so far out in the
- tails are neither meaningful, nor comparable for different degrees
- of freedom.
- Walt Pirie
-
-
-
- I sent a version of my comment on the Chi-Square probability question
- privately, but the direction of the public thread leads me to post
- my point to the list.
-
- The original question dealt with how to determine which of various
- Chi-Square results were "more significant" when the probability
- levels associated with the statistics were all in the form .0000
- The Chi-Square values were all present, but the probabilities were
- indistinguishable.
-
- I can't recall the original wording of the request, but my impression
- was that the various Chi-Square values, each of which had different
- degrees of freedom, were associated with various models tested on
- the same data set. If that impression is correct, then differences
- among the Chi-Square values can be tested simply by subtracting the
- Chi Square values and evaluating the resulting difference on the
- degrees of freedom that result from substracting the larger
- degrees of freedom from the smaller degrees of freedom..
-
- Of course, if the different Chi Square values do not reflect
- different models attempted on the same data, then none of the
- above applies.
-
- Dr. Francis C. Dane, Associate Professor and Chair
- Department of Psychology, Mercer University
- Macon, GA 31207-0001 USA
- Bitnet: FDANE@UGA Internet: FDANE@UGA.CC.UGA.EDU
- Tel: (912) 752-2972
- Fax: (912) 752-2108
-
- |==================================|============================|
- | Walter R. Pirie | |
- | Department of Statistics | |
- | Virginia Tech | |
- | Blacksburg, VA 24061-0439 | |
- | | |
- | Tel. 703-231-5441 | |
- | Bitnet WLTPIRIE@VTFVM1 | |
- | Telnet WLTPIRIE@VTVM1.CC.VT.EDU | |
- | | |
- | Fax 703-231-3863 | |
- |==================================|============================|
-