home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: bit.listserv.stat-l
- Path: sparky!uunet!zaphod.mps.ohio-state.edu!sdd.hp.com!news.cs.indiana.edu!lynx!nmsu.edu!dante!dclason
- From: dclason@dante.nmsu.edu (Dennis Clason)
- Subject: Re: tests for variance homogeneity
- Message-ID: <1992Aug26.201232.27843@nmsu.edu>
- Sender: usenet@nmsu.edu
- Organization: New Mexico State University, Las Cruces, NM
- References: <STAT-L%92082512373559@VM1.MCGILL.CA>
- Date: Wed, 26 Aug 1992 20:12:32 GMT
- Lines: 32
-
- In article <STAT-L%92082512373559@VM1.MCGILL.CA> BSAB604@UTXVM.BITNET writes:
- >Stat-L'ers:
- >
- > I am analyzing data from a three-factor field experiment
- >with replication. The residuals from the ANOVA model (computed
- >with SAS's PROC GLM) seem fairly well-behaved in my Q-Q plot.
- >Now, I need to assess the validity of the homogeneous variance
- >assumption.
- >
- God, I hate effects models! If you think about the
- experiment as a single classification (one-way) with
- a*b*c cells defined by the treatment combinations,
- it's clear that (given usual assumptions) the sufficient
- statistics for the problem are the cell means and
- variances.
-
- The short answer is Yes, you can use Levene's test.
- You should probably NOT use Bartlett's test or Hartley's
- F-max, neither being very robust (to non-normality).
-
- > What is the best approach to testing the assumption of
- >homogeneous variance? Does anyone use Levene's test in these
- >more complicated models? The SAS code to perform Levene's test
- >is trivial for one-way ANOVA...but can it be used on a model
- >like this?
- >
-
- Use the same SAS code, just set the model up as
- CLASS A B C;
- MODEL response = A*B*C / noint;
-
-
-