Manova outlier cutoff

I have been reading that Manova is very sensitive to outliers, particularly for small groups. One text I read suggests using 3.5 SD as a cutoff for outliers, but only deals with larger data sets. I wonder if it is acceptable to use more stingent cutoffs (2.5 SD) for uni and multivariate outliers when there is a smaller n, but am having a VERY difficult time finding firm guidelines.

Does anyone know the guidelines for outlier cutoffs for small vs larger groups, and if so, where can I find these?

For me, as a reviewer, the main point is what % of the data. You can use +/- 1SD for all I care if what is left accounts for 98% of the data. That's the real point ...

So, if you're losing 10% of the data ... that would be a big no-no for me. Anything after 2% sends warning bells ... and anything after 5% is like a war zone ...

Thanks, that does help a lot. I had not considered looking at the problem from a percentage of missing data perspective. To clarify, does 98% intact data incude data across all of the dependent variables for the MANOVA, or would it refer to 98% on a single variable?
Also, the purpose of removing outliers is of course different and dependent upon the types of conclusions you are attempting to derive/explain. However, regardless if you go with 2 1/2 or 3+ SDs you should be able to explain why you chose the cutoff points you have. Perhaps there is a specific type of occurrence that you are looking? For example, assuming a normal distribution, you only want those who are "normal" and show no trace of craziness on the Crazy Scale. Therefore you would want to go with say 1-2 sds above and below. But, the most important aspect (to me at least) is to be able to explain why you chose the cutoffs you chose. :yup:

Now that being said, i am not a reviewer and I would certainly trust the advice of Philyuko as he seems to be more experienced :tup:
Thanks to both Philyuko and Danielkeeton for your responses. You have both given me a new outlook on the problem (and reduced my stress level, which is always a good thing!) :)