ABSTRACT
The multivariate methods in this chapter deal with situations where two or more outcome variables are of interest. When dealing with multivariate data, the basic issue of detecting outliers might seem trivial: apply an outlier detection rule to each of the variables under study. One strategy for detecting outliers among multivariate data is to rotate the points in a variety of ways, each time using a boxplot to check for outliers among the marginal distributions. For the multivariate case, there is a classic method for testing the hypothesis that a measure of location based on means, called Hotelling's T2 test. The standard way of describing multivariate regression suggests that it takes into account the dependence among the outcome variables. The estimation procedure is tantamount to simply applying the least squares estimator to each of the dependent variables. Another approach to multivariate regression is to simply apply some robust estimator to each of the outcome variables under study.
