I assume you are estimating your model via GLM (Analyze > General Linear
Model > Repeated Measures via the GUI).  That would explain why you are
unable to omit the interaction term:  GLM includes it automatically.

To have more flexibility, I suggest you restructure your data from WIDE to LONG (look up VARSTOCASES, and see the examples).  Then use MIXED to  estimate your model.  The Command Syntax Reference manual (aka the FM)  includes an example of how to perform repeated measures ANOVA via MIXED.

Bruce Weaver

II assume that you want to do this because of the arguments made by Herbert Clark that both subjects and stimuli (words) should be considered random effects.  In traditional analyses, such as repeated measures ANOVA, subjects are a random effects factor and words are considered fixed effects. Because of limitation of the software of the time, to get words to be random effect, one had to "flip" the dataset so that the words became the unit of analysis.  I believe that the relevant publication for this is Clark 1973 which can be accessed here:
http://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=2&cad=rja&uact=8&ved=0CDEQFjAB&url=http%3A%2F%2Fcseweb.ucsd.edu%2F~gary%2FPAPER-SUGGESTIONS%2Fclark-jvlvb-1973.pdf&ei=UVxOU-H3BvOnsATPpYGIBQ&usg=AFQjCNFYlTLJX85pIjakWlwJbvN5QMaZSQ&sig2=IPqYrzb_Mqa9PdopjZwngw
(If Url get broken search scholar.google.com for "Clark", "1973" and "language as fixed-effect-fallacy".

However, this was a controversial assertion which not everyone in the psycholinguistics and memory research area agreed to.  You may want to see who cite this article and whether they are pro or con.

With newer software, factors in ANOVA can be defined as fixed (default) or random.  You may want to take a look at the following reference (Note: I got it from scholar.google.com and it is supposed to be in APA format but scholar doesn't get it right):

Carson, R. J., & Beeson, C. M. (2013). Crossing language barriers:Using crossed random effects modelling in psycholinguistics research. Tutorials Quant Meth Psych, 9(1), 25-41.The article is available here:www.tqmp.org/Content/vol09-1/p025/p025.pdf

This paper by Carson & Beeson (2013) provides background, rationale for doing an analysis where both subjects and words/stimuli are considered random effects, and provide SPSS examples using the MIXED procedure.

I never cared for Clark's argument but other have.  There are people on SPSS who are unfamiliar with Clark's argument but are SPSS guru's and can provide advice on to do the analysis presented by Carson & Beeson.
Mike Palij

Dear Mike,
yes, you are right, I would like to do an analysis by participants and stimuli,

but the main problem is that I can not find a way to switch off the interaction-calculating procedure in repeated measures Anova. This would be important, because the analysis by participant show a tendency
to be significant (p levels at 0.11 and 0.14) and the interaction is zero, so it was suggested to me to do the analysis without the interaction.

Thanks for the articles, I believe that I would not be confident in doing  the crossed-random effects model analysis

Here's another article on using multilevel models as an alternative to the old "Clarkian" methods. http://link.springer.com/article/10.3758/BF03192962.The Appendix includes both SAS and SPSS examples.

While looking for that link, I also found this "technical report", which anyone can download (no university library access required).  I've not read it, but have heard of the author*, and would guess it is probably fairly sound.

http://crr.ugent.be/papers/The%20language%20as%20fixed%20effect%20fallacy%20Version%202%200.pdf


* Years ago, I used a very nice Turbo Pascal timer routine Brysbaert published in Behavior Research Methods, Instruments & Computers.  Those who are interested in a little history can view it here: http://link.springer.com/article/10.3758/BF03209826.