6PAM 1036 - Multivariate Statistics CW2 Multivariate Statistics - CW2/Tutorial Sheet 4 Multivariate T-tests (To be submitted online by 11 March) Question 1. a) Consider an experiment which measures the value of three different vari- ables on a unit. This experiment was conducted twice, giving two samples (both with 22 observations each), with sample means 31 = 12= - 2 and sample covariance matrices 0 0 S 1 = 030 $ 2 = 0 2 0 0 5 i) Calculate the pooled sample covariance Spot for this data; [2 marks] ii) Calculate the corresponding Hotelling's T2-statistic and thus, conclude if the null hypothesis Ho should be rejected at the 1 significance level by comparison to a critical value from an F-table. [6 marks] b) If the null hypothesis in the two-sample T2-test is rejected, i.e. the two population means are not equal, we can determine which variable contributed the most to this rejection by finding the linear transformation coefficient vector a, which maximises the T-statistic T = am - a yz nithzaSpa nin2 where Spat is the pooled sample covariance. It can be shown that the coefficient vector which maximises this statistic is the so-called 'discriminant function a = Spill (31 - 12) . Using the discriminant function, show that the square of the maximised T-statistic is nothing other than the original Hotelling's T2 statistic for two samples, i.e. 12 = (31 - 32)(7 + 7 5 p) (31 -52). (7 marks]