Help with statistical analyses with only small sample size and multiple DV's??


New Member
I am looking at the relationship between Schizophrenia, Substance Abuse and Neurocognition.

I have 30 P's with Schizophrenia and 30 matched-controls (who are matched on age, gender, education level, IQ, and level of substance use), who have been administered a neurocognitive battery of 11 subtests. I want to look at the combined/additive effects of Schizophrenia and Substance Use on Neurocognitive performance.

Unfortunately, due to issues with recruitment and time constraints only small samples sizes could be achieved.

Is it possible to run more complex statistical analyses (other than just correlational analyses), despite the small sample sizes and multiple DV's??

If so, what is recommended to explore this relationship?


TS Contributor
If level of substance use was interval scaled you could run a model
with group, level of substance use, and their interaction as predictors.
Wheter you should run univariate analyses (e.g. multiple linear regressions)
or multivariate analyses (MANOVA) depends on the relationships of
your performance measures. If you assume that several measures
jointly represent one construct, these measures could be entered into

With kind regards



New Member
Hi K,

Thank you so much for your response. Level of substance abuse is measured by Frequency (Categorical: Low= 1-2 times a week, High= 3-7 times a week), Severity (grams), Chronicity (Years) and Time since last use (Years). All of the measures used in the neurocognitive battery are assumed to jointly represent one construct overall. However, they are also thought to represent seperate functions, so their individual relationship with the IV will be interesting to look at. I agree that a MANOVA seems to be the most appropriate statistical analyses to examine this relationship. However, is it possible to enter in all 11 measures in a MANOVA despite the small sample size? I have read that MANOVA's don't work well with a high number of DV's, particularly when there is a small sample size involved.

Thanks again,