# Sample size and documentation for discriminant analysis

Does anybody have good documentation for discriminant analysis? I have 9 variables (measurements), 60 patients and my outcome is good surgery, bad surgery. Also, is my sample size too small? Thank you

I have 3 points to say. First, what is "good" documentation for you? There is a lot of pages on DA and downloadable books on multivariate statistics (where DA is discussed) in the Internet. Some texts are superficial and easy, some are more sophisticated.

Second, because you have just 2 groups your DA will be virtually equivalent to the multiple linear regression with "Groups" dependent variable and your 9 variables as independent variables. The results (coefficients) after DA and the regression will be proportional between the two analyses. So, if you know linear regression that will suffice to understand the results even if you don't know DA.

Third, whenever there is just two groups it is worth considering to prefer binary logistic regression to DA. The logistic regression is less hard-to-plese in regard to assumptions than DA is. Just to mention some: it doesn't require multivariate normal distribution; it doesn't require equality of variances-covariances between the groups; it is less sensitive to disproportion of groups sizes as well as to outliers.

• Thank you. I am advising a student and he is interested in doing DA instead of logistic regression. I explained to him that I prefer logistic regression in his case and I will reiterate that to him. As for good documentation, I am looking for something that explains the discriminant functions and their coefficients as well as the canonical correlations and their meanings. I hope you can help.
– Sara
Commented Dec 20, 2011 at 11:43
• Can you also tell me when is DA preferred to logistic regression?
– Sara
Commented Dec 20, 2011 at 12:24
• @Sara, I'd explain fundamentals of DA right in my answer if I had time. Sorry. When assumptions of DA are met (i.e. normal distribution etc.) DA usually classifies objects somewhat more accurately than LR. Commented Dec 20, 2011 at 12:32