You can do a multipel regression analysis and ask how well treatment A and treatment B predict your DV. You can also ask in which session your DV works well . Then you get automatically Mahalonobis and Cook's test under the save button in SPSS. If you do a MANOVA you also control for outliers with Mahalanobis by pressing the save button SPSS. I think I would do a MANOVA in this case.

Often people search for outliers to eliminate them from the data. This is a bad practice that biases the outcome unless one can return to the original observation and show that the outlier was some sort of mistake. Removing errors is good, discarding an observation because it does not fit nicely is bad. What sort of error would you be able to find for a multivariate outlier that would justify its removal from the analysis? That said, it is perfectly reasonable to try the analysis with and without outliers and then discuss how conclusions change depending on the presence of these outliers. If there is little difference then the right approach is to use all of the data.

Just for the heck of it I actually studied this for part of my disertation. It was published and the reprint is attached. Good luck with your research. BTW this is for discriminate analysis but using a regression approach you can easily switch this to other linear model forms. See Snedecor and Cochran, STATISTICAL Methods try Google search or Amazon.com

Best wishes David Booth

I agree with Mr Timothy A Ebert. Removing errors is good, discarding an observation because it does not fit nicely is bad. But some time the value are highly anomalous from the average results like more than 3 times of the average results. it indicate there is an error in taking that particular observation. I think in such cases, conclusion of results considering outlier is not wrong.

Thank you, everyone!

The only purpose of this question was to get help in choosing a test to check the assumptions of ANOVA, before performing one. However, the discussion here seems to be misleading my question as to whether to remove the outliers or not.

NO, I'm not trying to remove outliers.

I just wanted to screen my data for outliers, hence I wanted to know what test could I use if I have multivariate data at different levels?

Thanks for your time and comments again!