Clusteranalysis
- 2. Cluster analysis is a group of multivariate techniques whose primary purpose is to group objects (e.g., respondents, products, or other entities) based on the characteristics they possess. It is a means of grouping records based upon attributes that make them similar. If plotted geometrically, the objects within the clusters will be close together, while the distance between clusters will be farther apart. * Cluster Variate - represents a mathematical representation of the selected set of variables which compares the object’s similarities.
- 3. Cluster Analysis - grouping is based on the distance (proximity) Factor Analysis -grouping is based on patterns of variation (correlation) Factor analysis, we form group of variables based on the several people’s responses to those variables. In contrast to Cluster analysis, we group people based on their responses to several variables.
- 4. Cluster Analysis - you don’t know who or what belongs to which group. Not even the number of groups. Discriminant analysis - requires you to know group membership for the cases used to derive classification rule.
- 5. Field of psychiatry - where the characterization of patients on the basis of clusters of symptoms can be useful in the identification of an appropriate form of therapy. Biology - used to find groups of genes that have similar functions. Information Retrieval - The world Wide Web consists of billions of Web pages, and the results of a query to a search engine can return thousands of pages. Clustering can be used to group these search results into small number of clusters, each of which captures a particular aspect of the query. For instance, a query of “movie” might return Web pages grouped into categories such as reviews, trailers, stars and theaters. Each category (cluster) can be broken into subcategories (sub-clusters_, producing a hierarchical structure that further assists a user’s exploration of the query results. Climate - Understanding the Earth’s climate requires finding patterns in the atmosphere and ocean. To that end, cluster analysis has been applied to find patterns in the atmospheric pressure of polar regions and areas of the ocean that have a significant impact on land climate.
- 6. Data Reduction -A researcher may be faced with a large number of observations that can be meaningless unless classified into manageable groups. CA can perform this data reduction procedure objectively by reducing the info. from an entire population of sample to info. about specific groups. Hypothesis Generation - Cluster analysis is also useful when a researcher wishes to develop hypotheses concerning the nature of the data or to examine previously stated hypotheses.
- 7. Cluster analysis is descriptive, atheoretical, and noninferential. Cluster analysis has no statistical basis upon which to draw inferences from a sample to a population, and many contend that it is only an exploratory technique. Nothing guarantees unique solutions, because the cluster membership for any number of solutions is dependent upon many elements of the procedure, and many different solutions can be obtained by varying one or more elements. Cluster analysis will always create clusters, regardless of the actual existence of any structure in the data. When using cluster analysis, the researcher is making an assumption of some structure among the objects. The researcher should always remember that just because clusters can be found does not validate their existence. Only with strong conceptual support and then validation are the clusters potentially meaningful and relevant. The cluster solution is not generalizable because it is totally dependent upon the variables used as the basis for the similarity measure. This criticism can be made against any statistical technique, but cluster analysis is generally considered more dependent on the measures used to characterize the objects than other multivariate techniques. With the cluster variate completely specified by the researcher. As a result, the researcher must be especially cognizant of the variables used in the analysis, ensuring that they have strong conceptual support.
- 8. Cluster analysis used for: ◦ Taxonomy Description. Identifying groups within the data ◦ Data Simplication. The ability to analyze groups of similar observations instead all individual observation. ◦ Relationship Identification. The simplified structure from CA portrays relationships not revealed otherwise. Theoretical, conceptual and practical considerations must be observed when selecting clustering variables for CA: ◦ Only variables that relate specifically to objectives of the CA are included. ◦ Variables selected characterize the individuals (objects) being clustered.
- 9. The primary objective of cluster analysis is to define the structure of the data by placing the most similar observations into groups. To accomplish this task, we must address three basic questions: ◦ How do we measure similarity? ◦ How do we form clusters? ◦ How many groups do we form?
- 10. Similarity represents the degree of correspondence among objects across all of the characteristics used in the analysis. It is a set of rules that serve as criteria for grouping or separating items. ◦ Correlational measures. - Less frequently used, where large values of r’s do indicate similarity ◦ Distance Measures. Most often used as a measure of similarity, with higher values representing greater dissimilarity (distance between cases), not similarity.
- 11. Both graph have the same r = 1, which implies to have a same pattern. But the distances (d’s) are not equal. 0 1 2 3 4 5 6 7 Category 1 Category 2 Category 3 Category 4 Chart Title 0 1 2 3 4 5 6 7 Category 1 Category 2 Category 3 Category 4 Graph 1 Graph 2 Graph 1 represents higher level of similarity
- 12. - several distance measures are available, each with specific characteristics. Euclidean distance. The most commonly recognized to as straight- line distance. Squared Euclidean distance. The sum of the squared differences without taking the square root. City- block (Manhattan) distance. Euclidean distance. Uses the sum of the variables’ absolute differences
- 13. Chebychev distance. Is the maximum of the absolute difference in the clustering variables’ values. Frequently used when working with metric (or ordinal) data. Mahalanobis distance (D2). Is a generalized distance measure that accounts for the correlations among variables in a way that weights each variables equally.
- 15. Suppose a marketing researcher wishes to determine market segments in a community based on patterns of loyalty to brands and stores a small sample of seven respondents is selected as a pilot test of how cluster analysis is applied. Two measures of loyalty- V1(store loyalty) and V2(brand loyalty)- were measured for each respondents on 0-10 scale.
- 17. Observation Observations A B C D E F G A --- B 3.162 --- C 5.099 2.000 --- D 5.099 2.828 2.000 --- E 5.000 2.236 2.236 4.123 --- F 6.403 3.606 3.000 5.000 1.414 --- G 3.606 2.236 3.606 5.000 2.000 3.16 2 --- Proximity Matrix of Euclidean Distance Between Observations
- 18. SIMPLE RULE: ◦ Identify the two most similar(closest) observations not already in the same cluster and combine them. ◦ We apply this rule repeatedly to generate a number of cluster solutions, starting with each observation as its own “cluster” and then combining two clusters at a time until all observations are in a single cluster. This process is termed a hierarchical procedure because it moves in a stepwise fashion to form an entire range of cluster solutions. It is also an agglomerative method because clusters are formed by combining existing clusters
- 19. AGGLOMERATIVE PROCESS CLUSTER SOLUTION Step Minimum Distance Unclustered Observationsa Observation Pair Cluster Membership Number of Clusters Overall Similarity Measure (Average Within-Cluster Distance) Initial Solution (A)(B)(C)(D)(E)(F)(G) 7 0 1 1.414 E-F (A)(B)(C)(D)(E-F)(G) 6 1.414 2 2.000 E-G (A)(B)(C)(D)(E-F-G) 5 2.192 3 2.000 C-D (A)(B)(C-D)(E-F-G) 4 2.144 4 2.000 B-C (A)(B-C-D)(E-F-G) 3 2.234 5 2.236 B-E (A)(B-C-D-E-F-G) 2 2.896 6 3.162 A-B (A-B-C-D-E-F-G) 1 3.420 In steps 1,2,3 and 4, the OSM does not change substantially, which indicates that we are forming other clusters with essentially the same heterogeneity of the existing clusters. When we get to step 5, we see a large increase. This indicates that joining clusters (B-C-D) and (E-F-G) resulted a single cluster that was markedly less homogenous.
- 20. Therefore, the three – cluster solution of Step 4 seems the most appropriate for a final cluster solution, with two equally sized clusters, (B-C-D) and (E-F-G), and a single outlying observation (A). This approach is particularly useful in identifying outliers, such as Observation A. It also depicts the relative size of varying clusters, although it becomes unwieldy when the number of observations increases.
- 22. Dendogram - Graphical representation (tree graph) of the results of a hierarchical procedure. Starting with each object as a separate cluster, the dendogram shows graphically how the clusters are combined at each step of the procedure until all are contained in a single cluster
- 23. Outliers can severely distort the representativeness of the results if they appear as structure (clusters) inconsistent with the objectives. ◦ They should be removed if the outliers represents: Abberant observations not representative of the population Observations of small or insignificant segments within the population and of no interest to the research objectives ◦ They should be retained if a undersampling/poor representation of relevant groups in the population; the sample should be augmented to ensure representation of these group.
- 24. Outliers can be identified based on the similarity measure by: ◦ Finding observations with large distances from all other observations. ◦ Graphic profile diagrams highlighting outlying cases. ◦ Their appearance in cluster solutions as single – member or small clusters.
- 25. The researcher should ensure that the sample size is large enough to provide sufficient representation of all relevant groups of the population The researcher must therefore be confident that the obtained sample is representative of the population.
- 26. Clustering variables that have scales using widely differing numbers of scale points or that exhibit large differences in standard deviations should de standardized. ◦ The most common standardization conversion is Z score (with mean equals to 0 and standard deviation of 1).
- 27. There are number of different methods that can be used to carry out a cluster analysis; these methods can be classified as follows: Hierarchical Cluster Analysis Nonhierarchical Cluster Analysis Combination of Both Methods
- 29. The stepwise procedure attempts to identify relatively homogeneous groups of cases based on selected characteristics using an algorithm either agglomerative or divisive, resulting to a construction of a hierarchy or treelike structure (dendogram) depicting the formation of clusters. This is one of the most straightforward method. HCA are preferred when: ◦ The sample size is moderate (under 300 – 400, not exceeding 1000).
- 30. Agglomerative Algorithm Divisive Algorithm *Algorithm- defines how similarity is defined between multiple – member clusters in the clustering process.
- 31. Hierarchical procedure that begins with each object or observation in a separate cluster. In each subsequent step, the two clusters that are most similar are combined to build a new aggregate cluster. The process is repeated until all objects a finally combined into a single clusters. From n clusters to 1. Similarity decreases during successive steps. Clusters can’t be split.
- 32. Begins with all objects in single cluster, which is then divided at each step into two additional clusters that contain the most dissimilar objects. The single cluster is divided into two clusters, then one of these clusters is split for a total of three clusters. This continues until all observations are in a single – member clusters. From 1 cluster to n sub clusters
- 34. Among numerous approaches, the five most popular agglomerative algorithms are: ◦ Single – Linkage ◦ Complete – Linkage ◦ Average – Linkage ◦ Centroid Method ◦ Ward’s Method ◦ Mahalanobis Distance
- 35. Single – Linkage ◦ Also called the nearest – neighbor method, defines similarity between clusters as the shortest distance from any object in one cluster to any object in the other.
- 36. Complete Linkage ◦ Also known as the farthest – neighbor method. ◦ The oppositional approach to single linkage assumes that the distance between two clusters is based on the maximum distance between any two members in the two clusters.
- 37. Average Linkage The distance between two clusters is defined as the average distance between all pairs of the two clusters’ members
- 38. Centroid Method ◦ Cluster Centroids - are the mean values of the observation on the variables of the cluster. ◦ The distance between the two clusters equals the distance between the two centroids.
- 39. Ward’s Method ◦ The similarity between two clusters is the sum of squares within the clusters summed over all variables. ◦ Ward's method tends to join clusters with a small number of observations, and it is strongly biased toward producing clusters with the same shape and with roughly the same number of observations.
- 40. The Hierarchical Cluster Analysis provides an excellent framework with which to compare any set of cluster solutions. This method helps in judging how many clusters should be retained or considered.
- 42. In contrast to Hierarchical Method, the NCA do not involve the treelike construction process. Instead, they assign objects into clusters once the number of clusters is specified. ◦ Two steps in Non HCA 1.) Specify Cluster Seed – identify starting points 2.) Assignment – assign each observation to one of the cluster seeds.
- 43. • Sequential Threshold Method • Parallel Threshold Method • Optimizing Procedures All of this belongs to a group of clustering algorithm known as K – means. ◦ K – means Method This method aims to partition n observation into k clusters in which each observation belongs to the cluster with the nearest mean. K – means is so commonly used that the term is used by some to refer to Nonhierarchical cluster analysis in general.
- 44. Simplicity. With the development of Dendogram, the HCA so afford the researcher with a simple, yet comprehensive portrayal of clustering solutions. Measures of Similarity. HCA can be applied to almost any type of research question. Speed. HCA have the advantage of generating an entire set of clustering solutions in an expedient manner.
- 45. To reduce the impact of outliers, the researcher may wish to cluster analyze the data several times, each time deleting problem observations or outliers. Hierarchical Cluster Analysis is not amenable to analyze large samples.
- 46. The results are less susceptible to outliers in the data, the distance measure used, and the inclusion of irrelevant or inappropriate variables. Non Hierarchical Cluster Analysis can analyze extremely large data sets.
- 47. Even a nonrandom starting solution does not guarantee an optimal clustering of observations. In fact, in many instances, the researcher will get a different final solution for each set of specified seed points. How is the researcher to select the optimum answer? Only by analysis and validation can the researcher select what is considered the best representation of structure, realizing that many alternatives may be acceptable. Nonhierarchical methods are also not so efficient when a large number of potential cluster solutions. Each cluster solution is a separate analysis, in contrast to the hierarchical techniques that generate all possible cluster solutions in a single analysis. Thus, nonhierarchical techniques are not as well suited to exploring a wide range of solutions based on varying elements such as similarity measures, observations included, and potential seed points.
- 48. A combination approach using a hierarchical approach followed by a nonhierarchical approach is often advisable. • First, a hierarchical technique is used to select the number of clusters and profile clusters centers that serve as initial cluster seeds in the nonhierarchical procedure. • A nonhierarchical method then clusters all observations using the seed points to provide more accurate cluster memberships. In this way, the advantages of hierarchical methods are complemented by the ability of the nonhierarchical methods to refine the results by allowing the switching of cluster membership.
- 49. The cluster centroid, a mean profile of the cluster on each clustering variable, is particularly useful in the interpretation stage: ◦ Interpretation involves examining the distinguishing characteristics of each cluster’s profile and identifying substantial differences between clusters. ◦ Cluster solutions failing to show substantial variation indicate other cluster solutions should be examined. ◦ The cluster centroid should also be assessed for correspondence with the researcher’s prior expectations based on theory or practical experience.
- 50. Validation is essential in cluster analysis because the clusters are descriptive of structure and require additional support for their relevance: • Cross – validation empirically validates a cluster solution by creating two subsamples (randomly splitting the sample) and then comparing the two cluster solutions for consistency with respect to the number of clusters and the clusters profiles. • Validation is also achieved by examining differences on variables not included in the cluster analysis but for which a theoretical and relevant reason enables the expectation of variation across the clusters.
- 51. Cluster Analysis on SPSS
- 52. Hierarchical Cluster Analysis Non Hierarchical Cluster Analysis Two – Step Cluster Analysis
- 54. Example: This file only includes 20 cases, each responding to items on demographics (gender, qualifications, days absence from work, whether they smoke or not), on their attitudes to smoking in public places (subtest totals for pro and anti), plus total scale score for self-concept. We are attempting to determine how many natural groups exist and who belongs to each group.
- 55. The initial step is determining how many groups exist. The SPSS hierarchical analysis actually calculates every possibility between everyone forming their own group (as many clusters as there are cases) and everyone belonging to the same group, giving a range in our data of from 1 to 20 clusters.
- 56. Click ‘Plots’
- 57. Click ‘Method’ Select Continue then OK.
- 58. Parsing the classification tree to determine the number of clusters is a subjective process. Generally, you begin by looking for "gaps" between joining along the horizontal axis. Starting from the right, there is a gap between 5 and 10, which splits the data into two clear clusters and a minor one. DENDOGRAM
- 59. AGLOMERATION SCHEDULE The agglomeration schedule is a numerical summary of the cluster solution. At the first stage, cases 7 and 9 are combined because they have the smallest distance The cluster created by their joining next appears in stage 8 In stage 8, the observations 5 and 7 were joined. The resulting cluster next appears in stage 13. When there are many cases, this table becomes rather long, but it may be easier to scan the coefficients column for large gaps rather than scan the dendrogram.
- 60. A good cluster solution sees a sudden jump (gap) in the distance coefficient. The solution before the gap indicates the good solution. Table 23.2 is a reformed table to see the changes in the coefficients as the number of clusters increase. The final column, headed 'Change’, enables us to determine the optimum number of clusters. In this case it is 3 clusters as succeeding clustering adds very much less to distinguishing between cases.
- 61. Repeat step 1 to 3 to place cases into one of three clusters. The number you place in the box is the number of clusters that seem best to represent the clustering solution in a parsimonious way. Finally click OK.
- 62. A new variable has been generated at the end of your SPSS data file called clu3_1 (labelled Ward method in variable view). This provides the cluster membership for each case in your sample Nine respondents have been classified in cluster 2, while there are seven in cluster 1 and four in cluster 3.
- 63. We now proceed by conducting a one-way ANOVA to determine on which classifying variables are significantly different between the groups
- 64. Qualifications did not produce any significant associations.
- 65. Crosstab analysis of the nominal variables gender, qualifications and whether smoke or not produced some significant associations with clusters.
- 66. Cluster 1 : consists of male non smokers Cluster 2 : consists of smoking and non smoking males Cluster 3 : consists of smoking females
- 67. Cluster 1 is characterized by low self-concept, average absence rate, average attitude score to anti-smoking, non-smoking males. Cluster 2 is characterized by moderate self-concept, low absence rate, average attitude score to anti-smoking, smoking and non-smoking males. Cluster 3 is characterized by high self-concept, high absence rate, low score to antismoking, smoking females.
- 69. Example: The telecommunication provider wants to segment its customer base by service usage patterns. If customer can be classified by usage, the company can offer more attractive package to its customers. The following variables are; Multiple lines Voice mail Paging service Internet Caller ID Call waiting Call forwarding 3-way calling Electronic billing
- 71. Specify number of clusters Number iteration or repetition of combining different clusters. Determines when iteration cease and represent a proportion of the min. distance bet. Initial cluster center.
- 72. STATISTICS it will show the information for each group.
- 73. The initial cluster centers are the variable values of the k well- spaced observations.
- 74. In early iterations, the cluster centers shift quite a lot. Iteration History The iteration history shows the progress of the clustering process at each step. By the 14th iteration, they have settled down to the general area of their final location, and the last four iterations are minor adjustments.
- 75. ANOVA The ANOVA table indicates which variables contribute the most to your cluster solution. Variables with large F values provide the greatest separation between clusters.
- 76. FINAL CLUSTER CENTERS The final cluster centers are computed as the mean for each variable within each final cluster. The final cluster centers reflect the characteristics of the typical case for each cluster. Customers in cluster 1 tend to be big spenders who purchase a lot of services.
- 77. FINAL CLUSTER CENTERS Customers in cluster 2 tend to be moderate spenders who purchase the "calling" services. Customers in cluster 3 tend to spend very little and do not purchase many services.
- 78. This table shows the Euclidean distances between the final cluster centers. Greater distances between clusters correspond to greater dissimilarities. Clusters 1 and 3 are most different. DISTANCES BETWEEN FINAL CLUSTER CENTER Cluster 2 is approximately equally similar to clusters 1 and 3.
- 79. Number of Cases in Each Cluster A large number of cases were assigned to the third cluster, which unfortunately is the least profitable group. Perhaps a fourth, more profitable, cluster could be extracted from this "basic service" group.
- 80. MAIN DIALOG BOX
- 81. SAVE DIALOG BOX Creates a new variable indicating the final cluster membership of each case. Creates a new variable indicating the Euclidean distances bet. Each cases and its classification center.
- 82. FINAL CLUSTER CENTERS This table shows that an important grouping is missed in the three-cluster solution Members of clusters 1 and 2 are largely drawn from cluster 3 in the three-cluster solution, and they are unlikely to be big spenders. Clusters 3 and 4 seem to correspond to clusters 1 and 2 from the three- cluster solution.
- 83. DISTANCE BETWEEN FINAL CLUSTER CENTERS Clusters 1 and 2 are the most similar, which makes sense because they were combined into one cluster in the three-cluster solution. Cluster 4 is still equally similar to the other clusters.
- 84. NUMBER OF CASES IN EACH CLUSTER Nearly 25% of cases belong to the newly created group of "E-service" customers, which is very significant to your profits.
- 86. The Two Step Cluster Analysis procedure is an exploratory tool designed to reveal natural groupings (or clusters) within a data set that would otherwise not be apparent. The algorithm employed by this procedure has several desirable features that differentiate it from traditional clustering techniques: • The ability to create clusters based on both categorical and continuous variables. • Automatic selection of the number of clusters. • The ability to analyze large data files efficiently.
- 87. In order to handle categorical and continuous variables, the TwoStep Cluster Analysis procedure uses a likelihood distance measure which assumes that variables in the cluster model are independent. Further, each continuous variable is assumed to have a normal (Gaussian) distribution and each categorical variable is assumed to have a multinomial distribution. Empirical internal testing indicates that the procedure is fairly robust to violations of both the assumption of independence and the distributional assumptions, but you should try to be aware of how well these assumptions are met. The two steps of the TwoStep Cluster Analysis procedure's algorithm can be summarized as follows: Step 1. The procedure begins with the construction of a Cluster Features (CF) Tree. The tree begins by placing the first case at the root of the tree in a leaf node that contains variable information about that case. Each successive case is then added to an existing node or forms a new node, based upon its similarity to existing nodes and using the distance measure as the similarity criterion. A node that contains multiple cases contains a summary of variable information about those cases. Thus, the CF tree provides a capsule summary of the data file. Step 2. The leaf nodes of the CF tree are then grouped using an agglomerative clustering algorithm. The agglomerative clustering can be used to produce a range of solutions. To determine which number of clusters is "best", each of these cluster solutions is compared using Schwarz's Bayesian Criterion (BIC) or the Akaike Information Criterion (AIC) as the clustering criterion.
- 88. Car manufacturers need to be able to appraise the current market to determine the likely competition for their vehicles. If cars can be grouped according to available data, this task can be largely automatic using cluster analysis.