Contents Index <<Browse<< >>Browse>>
Here the basic features of using proc discrim for discriminant analysis are studied.
The Fisher Iris data contained in iris.dat, and examined in Project 3, is used again. The objective is to determine if the species of a given iris plant can be determined simply on the basis of the four measurements.
1. What statistical test should always be done before a discriminant analysis is attempted? WHY?
2. Assume the measurements for each species are a sample from a multivariate normal population. Assume also that the covariance matrices for the 3 populations are equal. Carry out a discriminant analysis using proc discrim. Is discrimination successful? What type of discrimination rule is used? Which observations are misclassified? Submit nice print outs from SAS to back up your answers. Turn in a nice print out of your program too. There's no need to print out the data.
3. Repeat the analysis of problem 1, but this time test (alpha=0.01) for equality of covariance matrices of the 3 populations. Is the discrimination rule different? If so, what is the discrimination rule now and how successful is it? Submit nice print outs from SAS to back up your answers.
4. Instead of using the full data set to devise a discrimination rule, use only the first half of the observations for each species. For simplicity, assume normality and equality of covariance matrices. What is the discrimination rule? Use this discrimination rule on the remaining half of the observations. What are your estimated error rates? How do these error rates compare to the estimated rates in problem 1? Submit nice print outs from SAS to back up your answers. Turn in a copy of your program too--but not the data itself.
Contents Index <<Browse<< >>Browse>>
Copyright © 1997 by Jerry Alan Veeh. All rights reserved.