ANOVA :unequal sample sizes
Mahmood Abolghasemi
I have three groups of different sizes : group one :20, group two: 53,
and group three 212. Can I use one-way analysis of variance to make
comparison between these groups? Any comments apprecited.
David Nichols
In article <51tlvb$n...@mirv.unsw.edu.au>,
SPSS procedures for analysis of variance do not assume equal sample
sizes. You can use ONEWAY, for example, to analyze such data. The
robustness properties of the F-test are not as good with unequal N
as with equal N.
--
-----------------------------------------------------------------------------
David Nichols Senior Support Statistician SPSS, Inc.
Phone: (312) 329-3684 Internet: nic...@spss.com Fax: (312) 329-3668
-----------------------------------------------------------------------------
Robert Virzi
Mahmood Abolghasemi <p212...@csdvax.csd.unsw.edu.au> wrote:
>>I have three groups of different sizes : group one :20, group two: 53,
>>and group three 212. Can I use one-way analysis of variance to make
>>comparison between these groups? Any comments apprecited.
David Nichols <nic...@spss.com> responded:
>SPSS procedures for analysis of variance do not assume equal sample
>sizes. You can use ONEWAY, for example, to analyze such data. The
>robustness properties of the F-test are not as good with unequal N
>as with equal N.
Hmmmm. What about for repeated measures anova? I seem to recall
a problem in a recent nXm repeated measures design I had to analyze
with SPSS (mac 6.1.1) because of dropped trials. I seem to recall
getting what was effectively listwise elimination of any case with
a missing variable. Substituting cell means fixes the problem, in
that the analysis can be run.
Did I miss an option that would have dealt with this problem for
repeated measures or mixed models?
-Bob Virzi
--
rvi...@gte.com Just another ascii character...
+1(617)466-2881
David Nichols
No, you haven't missed any options. I guess it's a matter of differences
in terminology here. You can have unequal numbers of cases in the cells
of your between subjects design, but having different numbers of variables
observed in a repeated measures is going to result in dropping any cases
without all time points. To me this is missing data as opposed to unequal
cell sizes. The best MANOVA can do with such data if you don't want to
use LISTWISE deletion is to give fairly simple mixed model results with
the data being set up with multiple cases per subject, an explicit subject
variable, and user assignment of error terms. The new GLM (General Linear
Model) procedure in release 7.0 will take this kind of data and allow you
to simply tell it that the subject factor is random, and it will compute
expected mean squares and assign error terms based on these (which in the
general missing time points case will produce quasi-F ratios).