1. Trang chủ
  2. » Luận Văn - Báo Cáo

Lecture Marketing research (12th edition) - Chapter 20: Discriminant, factor and cluster analysis

48 64 0

Đang tải... (xem toàn văn)

Tài liệu hạn chế xem trước, để xem đầy đủ mời bạn chọn Tải xuống

THÔNG TIN TÀI LIỆU

Thông tin cơ bản

Định dạng
Số trang 48
Dung lượng 816,23 KB

Nội dung

Chapter 20 - Discriminant, factor and cluster analysis. In this chapter, the following content will be discussed: Discriminant analysis, objectives of discriminant analysis, basic concept, discriminant function, discriminant function – a graphical illustration,...

1 Marketing Research Aaker, Kumar,  Leone and Day  Twelfth Edition Instructor’s  Chapter Twenty Discriminant, Factor and Cluster Analysis / Marketing Research 12th Edition Discriminant Analysis • • Used to classify individuals into one of two or more  alternative groups on the basis of a set of measurements Used to identify variables that discriminate between  naturally occurring groups Major Uses Prediction / Description Marketing Research 12th Edition Objectives of Discriminant  Analysis • • • • / Determining linear combinations of the predictor variables to  separate groups by measuring between­group variation relative to  within­group variation Developing procedures for assigning new objects, firms, or  individuals, whose profiles, but not group identity are known, to one  of the two groups Testing whether significant differences exist between the two groups  based on the group centroids Determining which variables count most in explaining inter­group  differences Marketing Research 12th Edition Basic Concept If we can assume that two populations have the same variance, then the usual value of C is where X1 and XII are the mean values for the two groups, respectively Distribution of two populations / Marketing Research 12th Edition Discriminant Function Zi = b1 X1 + b2 X2 + b3 X3 +   + bn Xn Where            Z  = discriminant score            b  = discriminant weights            X  = predictor (independent) variables      In a particular group, each individual has a discriminant score (zi) Σ zi = centroid (group mean); where i = individual Indicates most typical location of an individual from a  particular group / Marketing Research 12th Edition Discriminant Function –  A Graphical Illustration / Marketing Research 12th Edition Cut­off  Score • Criterion against which each individual’s discriminant score is  judged to determine into which group the individual should be  classified For equal group sizes / For unequal group sizes Marketing Research 12th Edition Determination of Significance • • • / Null Hypothesis: In the population, the group means the  discriminant function are equal Ho : μA = μB  Generally, predictors with relatively large standardized  coefficients contribute more to the discriminating power of  the function Canonical or discriminant loadings show the variance that  the predictor shares with the function Marketing Research 12th Edition 10 Classification and Validation Holdout Method • • • / Uses part of sample to construct classification rule; other subsample  used for validation Uses classification matrix and hit ratio to evaluate groups  classification Uses discriminant weights to generate discriminant scores for cases in  subsample Marketing Research 12th Edition 34 Hierarchical Clustering • Single Linkage ▫ • Complete Linkage ▫ / Clustering criterion based on  the shortest distance Clustering criterion based on  the longest distance Marketing Research 12th Edition 35 Hierarchical Clustering (Contd.) • Average Linkage ▫ Clustering criterion based  on the average distance • Ward's Method ▫ / Based on the loss of  information resulting from  grouping of the objects into  clusters (minimize within  cluster variation) Marketing Research 12th Edition 36 Hierarchical Clustering (Contd.) • Centroid Method ▫ Based on the distance between the group centroids  (the point whose coordinates are the means of all the  observations in the cluster) / Marketing Research 12th Edition 37 Hierarchical Cluster Analysis ­  Example / Marketing Research 12th Edition 38 Hierarchical Cluster Analysis  (Contd.) A dendrogram for hierarchical clustering of bank data / Marketing Research 12th Edition 39 Hierarchical Cluster Analysis  (Contd.) / Marketing Research 12th Edition 40 Non­hierarchical Clustering • Sequential Threshold  ▫ • Parallel Threshold ▫ • Several cluster centers are selected and objects within threshold level  are assigned to the nearest center Optimizing ▫ / Cluster center is selected and all objects within a pre­specified  threshold value are grouped Objects can be later reassigned to clusters on the basis of optimizing  some overall criterion measure Marketing Research 12th Edition 41 Nonhierarchical Cluster Analysis ­ Example / Marketing Research 12th Edition 42 Nonhierarchical Cluster  Analysis – Example (Contd.) / Marketing Research 12th Edition 43 Nonhierarchical Cluster  Analysis – Example (Contd.) / Marketing Research 12th Edition 44 Nonhierarchical Cluster  Analysis – Example (Contd.) / Marketing Research 12th Edition 45 Criteria for Determining the  Number of Clusters / ▫ Number  of  clusters  is  specified  by  the  analyst  for  theoretical  or  practical reasons ▫ Level of clustering with respect to clustering criterion is specified ▫ Determine the number of clusters from the pattern of clusters  generated. The distances between clusters or error variability measure  at successive steps can be used to decide the number of clusters (from  the plot of error sum of squares with the number of clusters) ▫ The ratio of total within­group variance to between group variance is  plotted against the number of clusters and the point at which an elbow  Marketing Research 12th Edition 46 Methods to Validate a Cluster  Analysis Solution  • • • • / Apply two or more different clustering approaches to same data or use different distance  measures and compare the results Split the data randomly into two halves and perform clustering on each half and then  examine the average profile values of each cluster across sub samples Delete various columns (variables) from the original data, compute dissimilarity measures  across remaining variables and compare these results with the results obtained using full  set Using simulation procedures create a data set with the properties matching the overall  properties of the original data but containing no clusters. Use the same clustering method  on both original and the artificial data and compare the results Marketing Research 12th Edition Assumptions and Limitations of  Cluster Analysis • 47 Assumptions The basic measure of similarity on which the clustering is based is a  valid measure of the similarity between the objects ▫ There is theoretical justification for structuring the objects into  clusters ▫ • Limitations It is difficult to evaluate the quality of the clustering ▫ It is difficult to know exactly which clusters are very similar and  which objects are difficult to assign ▫ It is difficult to select a clustering criterion and program on any  basis other than availability ▫ / Marketing Research 12th Edition 48 End of Chapter Twenty / Marketing Research 12th Edition ... Common Factor Analysis – Results (Contd.) / Marketing Research 12th Edition 29 Common Factor Analysis ­ Results / Marketing Research 12th Edition 30 Common Factor Analysis – Results (Contd.) / Marketing. ..2 Chapter Twenty Discriminant, Factor and Cluster Analysis / Marketing Research 12th Edition Discriminant Analysis • • Used to classify individuals into one of two or more ... to a smaller set of factors • Common Factor Analysis ▫ Uncovers underlying dimensions surrounding the  original variables / Marketing Research 12th Edition 19 Factor Analysis ­ Example / Marketing Research 12th Edition

Ngày đăng: 18/01/2020, 23:26

TỪ KHÓA LIÊN QUAN