Free Essay

Cluster analysis, like reduced space analysis (factor analysis), is concerned with data matrices in which the variables have not been partitioned beforehand into criterion versus predictor subsets. In reduced space analysis our interest centers on reducing the variable space to a smaller number of orthogonal dimensions, which maintains most of the information–metric or ordinal– contained in the original data matrix. Emphasis is placed on the variables rather than on the subjects (rows) of the data matrix. In contrast, cluster analysis is concerned with the similarity of the subjects–that is, the resemblance of their profiles over the whole set of variables. These variables may be the original set or may consist of a representation of them in reduced space (i.e., factor scores). In either case the objective of cluster analysis is to find similar groups of subjects, where “similarity” between each pair of subjects is usually construed to mean some global measure over the whole set of characteristics–either original variables or derived coordinates, if preceded by a reduced space analysis.

In this section we discuss various methods of clustering and the key role that distance functions play as measures of the proximity of pairs of points. We first discuss the fundamentals of cluster analysis in terms of major questions concerning choice of proximity measure, choice of clustering technique, and descriptive measures by which the resultant clusters can be defined. We show that clustering results can be sensitive to the type of distance function used to summarize proximity between pairs of profiles. We next discuss the characteristics of various computational algorithms that are used for grouping profiles, i.e., for partitioning the rows (subjects) of the data matrix. This is followed by brief discussions of statistics for defining clusters and the problems associated with statistical inference in this area.

Basic Questions in Cluster Analysis

The most common use of cluster analysis is classification. That is, subjects are separated into groups such that each subject is more similar to other subjects in its group than to subjects outside the group. Cluster analysis is thus concerned ultimately with classification and represents a set of techniques that are part of the field of numerical taxonomy

(Frank and Green, [1968]; Punj and Stewart [1983]; Aldenderfer and Blashfield [1984]).

We will initially focus on clustering procedures that result in the assignment of each subject to one and only one class.

Subjects within a class are usually assumed to be indistinguishable from one another. Thus, we assume that the underlying structure of the data involves an unordered set of discrete classes. In some cases we may also view these classes as hierarchical in nature, with some classes divided into subclasses.

Clustering procedures can be viewed as “pre-classificatory” in the sense that the researcher has not used prior judgment to partition the subjects (rows of the data matrix). However, it is assumed that some of the objectives are heterogeneous; that is, that “clusters” exist. This presupposition of different groups is based on commonalities within the set of independent variables. This assumption is different from that made in the case of discriminant analysis or automatic interaction

Nov 1, 2011 revision of Green, P.E., F. J. Carmone and S. M. Smith, Multidimensional Scaling, SECTION FIVE: DIMENSION REDUCING METHODS AND

CLUSTER ANALYSIS, Addison Wesley, 1989.

1

Cluster Analysis | 1

detection, where the dependent variable is used to formally define groups of objects and the distinction is not made on the basis of profile resemblance in the data matrix itself. Thus, given that no information on group definition is formally evaluated in advance, the major problems of cluster analysis will be discussed as follows:

1. What measure of inter-subject similarity is to be used and how is each variable to be “weighted” in the construction of such a summary measure?

2. After inter-subject similarities are obtained, how are the classes to be formed?

3. After the classes have been formed, what summary measures of each cluster are appropriate in a descriptive sense; that is, how are the clusters to be defined?

4. Assuming that adequate descriptions of the clusters can be obtained, what inferences can be drawn regarding their statistical significance?

Choice of Proximity Measure

The choice of proximity, similarity, association, or resemblance measure (all four terms will be used synonymously here) is an interesting problem in cluster analysis. The concept of similarity always connotes the question: similarity with respect to what? Proximity measures are usually viewed in relative terms–two objects are similar, relative to the group, if their profiles across variables are “close” or they share “many” aspects in common, relative to those which other pairs share in common. Most clustering procedures use pairwise measures of proximity. The choice of which subjects and variables to use in the first place is largely a matter for the researcher’s judgment. While these (prior) choices are important ones, they are beyond our scope of coverage. Even assuming that such choices have been made, however, the possible measures of pairwise proximity are many. Generally speaking, these measures fall into two classes: (a) distance-type measures (including correlation coefficients); and (b) matching-type measures. The characteristics of each class are discussed in turn.

Distance-Type Measures

A surprisingly large number of proximity measures–including correlation measures–can be viewed as distances in some type of metric space. In Section 2 we introduced the notion of Euclidean distance between two points in a space of r dimensions. We recall that the formula was:

where xij, xjk are the projections of points i and j on dimension k; (k = 1,2,…,r). In as much as the variables are often measured in different units, the above formula is usually applied after each variable has been standardized to mean zero and unit standard deviation. Our subsequent discussion will assume that this preliminary step has been taken. The Euclidean

Cluster Analysis | 2

distance measure assumes that the space of (standardized) variables is orthogonal, i.e., that the variables are uncorrelated. While the Euclidean measure can still be used with correlated variables, it is useful to point out that (implicit) weighting of the components underlying the associated variables occurs with the use of the Euclidean measure:

1. Squared Euclidean distance in the original variable space has the effect of weighting each underlying principal component by that component’s eigenvalue.

2. Squared Euclidean distance in the component space (where all components are first standardized to unit variance) has the effect of assigning equal weights to all components.

3. In terms of the geometry of the configuration, in the first case all points are rotated to orthogonal axes with no change in squared inter-point distance. The general effect is to portray the original configuration as a hyper- ellipsoid with principal components serving as axes of that figure. Equating all axes to equal length has the effect of transforming the hyper-ellipsoid into a hyper-sphere where all “axes” are of equal length.

The above considerations can be represented in terms of the following squared distance model:

where: yik, yjk denote unit variance components of profiles i and j on component axis k (k = 1,2,…,r). If one weights the component scores according to the variances of the components (before standardization) the expression is:

where

is the k-th component variance, or eigenvalue. This expression is equivalent to d2ij expressed in original variable space. The above relationships assume that all principal components are extracted. As described earlier, if such is not the case, squared inter-point distances will be affected by the fact that they are computed in a component space of lower dimensionality than the original variable space. In summary, both the Euclidean distance measure in original variable space and the Euclidean distance in component space (assuming all components have been extracted) preserve all of the information in the original data matrix. Finally it should be pointed out that if (in addition to being standardized to mean zero and unit variance) the original variables are uncorrelated, both d2ij and *d2ij will be equivalent.

Other Euclidean Distance Measures

Two other measures have often been proposed as proximity measures. Both of these measures derive from historical clustering methods, which used Q-type factor analysis to cluster subjects. In Q-type factor analysis–as described briefly in the Qualtrics White Paper on Factor Analysis–the correlation (or covariance) matrix to be factored consists of inter-subject rather than inter-variable proximities. In these methods the weights k are left intact.

Cluster Analysis | 3

Figures 1 and 2 show these effects geometrically. In the case of either covariance or correlation matrices the profile mean is subtracted from each vector component that, in the 2-component case of Figure 1, results in a centroid with an (new) origin located at point X on the figure. Figure 2 shows the effect of removing profile dispersion. If we assume that the profiles were originally positioned in three-space, removal of each profile’s mean reduces their dimensionality to two-space.

By using a correlation matrix we further reduce dimensionality by projecting all points on to the unit circle, since the distance of the profile point can represent the standard deviation of a profile from the origin (the centroid of the points first having been translated to the origin). Thus, profiles a, b, and c would all be identical after the transformation, as would profiles d and e.

The cosine of the angle separating the two vectors represents the Q-correlation between them. Of course, there may be cases where the researcher is not interested in profile differences due to either mean and/or dispersion. If so, a Q-type analysis applied to covariance or correlation matrices, as the case may be, is perfectly sensible even though information is (willingly) discarded.

Figure

1

Effect of Q-‐type Component Analysis on Profile Means

Figure 2

Effect of Q-‐type Component Analysis on Profile Dispersion

In general, we should expect differences in the derived squared distance measure computed from these procedures–both between themselves and between those computed by the techniques previously discussed. While the authors have a predilection for the information-preserving measures d2ij and *d2ij, it is well to point out that no universally applicable distance measure exists. The choice of which measure to use depends upon which aspect of the data is worth “preserving.”

A wide variety of distance-type measures are available for cluster analysis; several of which are compared by Aldenderfer and Blashfield (1984). Once the researcher has selected a method of measuring pairwise profile similarity, the computational routine for clustering the subjects must be selected. Aldenderfer and Blashfield identify several families of clustering methods, each of which uses a different approach to creating groups: (1) hierarchical agglomerative; (2) hierarchical divisive; (3) factor analytic; (4) non-hierarchical.

Hierarchial Methods

These procedures are characterized by the construction of a hierarchy or tree-like structure. In some methods each point starts out as a unit (single-point) cluster. At the next level the two closest points are placed in a cluster. At the next level a third point joins the first two or else a second two-point cluster is formed, based on various criterion rules for assignment.

Cluster Analysis | 4

In application, hierarchical clustering is useful in determining if points are substitutable rather than mutually exclusive.

The different assignment rules include:

SINGLE LINKAGE RULE

The single linkage or minimum distance rule starts out by finding the two points with the minimum distance. These are placed in the first cluster. At the next stage a third point joins the already-formed cluster of two if the minimum distance to any of the members of the cluster is smaller than the distance between the two closest unclustered points. Otherwise, the two closest unclustered points are placed in a cluster. The process continues until all points end up in one cluster. The distance between two clusters is defined as the shortest distance from a point in the first cluster that is closest to a point in the second.

COMPLETE LINKAGE RULE

The complete linkage option starts out in just the same way by clustering the two points with the minimum distance.

However, the criterion for joining points to clusters or clusters to clusters involves maximum (rather than minimum) distance. That is, a third point joins the already formed cluster of two if the maximum distance to any of the members of the cluster is smaller than the distance between the two closest unclustered points. In other words, the distance between two clusters is the longest distance from a point in the first cluster to a point in the second cluster.

AVERAGE LINKAGE

The average linkage option starts out in the same way as the other two. However, in this case the distance between two clusters is the average distance from points in the first cluster to points in the second cluster.

WARD’S METHOD

Ward’s Method starts out by finding two points with the minimum within groups sum of squares. Points continue to be joined to the first cluster or to other points depending on which combination minimizes the error sum of squares from the group centroid. This method is also known as a k-means approach. Closely related to Ward’s algorithm is the HowardHarris algorithm [Harris, 1981]. The Howard-Harris algorithm is a hierarchical divisive method which uses the k-means method of assigning cases to the clusters. The k-means method assigns the case to the closest centroid. The approach may take either of the two forms described below:

K-MEANS APPROACH #1

1. Initially the entire set of observations is considered as one set. The group is split based on the one variable which makes the greatest contribution to within-group sum of squares.

2. Group centroids are re-computed and subject distances to all group centroids are computed. The subject that

Cluster Analysis | 5

would best improve the objective function is re-assigned. This process is repeated until a finite number of transfers are performed, no further improvement in within-groups sum of squares is found, or a local optimum is reached.

3. The group with the largest within-groups sum of squares is selected for splitting. Steps 2 and 3 are then repeated until the desired number of clusters is identified.

K-MEANS APPROACH #2

1. An m x m covariance matrix is formed and analyzed using principal components analysis. A factor score is computed for each of the n subjects on the first (and most important) dimension or factor.

2. All subjects with factor scores that exceed the mean value of the factor are assigned into a new cluster.

3.

After splitting, each observation is re-evaluated against all clusters. If the objective function is improved by re- assigning a case to another cluster, the case making the greatest improvement is re-assigned. Optimization continues until a) finite number of transfers are performed, b) no further improvement in the objective function is found, or c) a local optimum is reached.

4. The next factor is selected as the basis for splitting the next cluster. Steps 2, 3, and 4 are then repeated until the desired number of clusters is identified.

Hierarchical clustering algorithms may be identified as either hierarchical agglomerative or hierarchical divisive, meaning that they contract or expand the space between groups of points in the multivariate space. Divisive linkage methods tend to start new clusters rather than join points to existing clusters. Ward’s method and complete linkage rules are of the divisive variety and tend to create clusters of roughly equal size that are hyper-spherical in form. The average linkage method neither expands nor contracts the original space, while the single linkage tends to agglomerate or contract the space between groups of points in multivariate space.

FACTOR ANALYTIC METHODS FOR CLUSTERING

These methods analyze an n x n correlation matrix of similarities between the n cases to find a dimensional representation of the points. Clusters are then developed based on the resulting factor loadings (the correlation between the subject and the underlying dimension). Clustering using factor analytic methods has been criticized because of the use of a linear model that is developed across cases rather than across variables. The linear model tends to moderate the correct identification of anything other than linear additive predictive groups.

NON HIERARCHICAL METHODS FOR CLUSTERING

Non Hierarchical Methods have in general been subject to limited use and testing, making their specific operational characteristics difficult to identify. In general, however, these methods start right from a proximity matrix and work in the

Cluster Analysis | 6

1. Begin with an initial split of the data into a specified number of clusters.

2. Allocate each data point to the cluster with the nearest centroid.

3. Compute the new centroid of the clusters after all points are assigned to clusters.

4. Iterate steps 2 and 3 until no further changes occur.

Several general types of non-hierarchical clustering designs exist and can be characterized:

• SEQUENTIAL THRESHOLD–in this case a cluster center is selected and all objects within a pre-specified threshold value are grouped. Then a new cluster center is selected and the process is completed. Once points enter a cluster they are removed from further processing.

• PARALLEL THRESHOLD–this method is similar to the one immediately above except that several cluster centers are selected in advance and points within the threshold level are assigned to the nearest center; threshold levels can then be adjusted to admit fewer or more points to clusters.

• PARALLEL PARTITIONING–this method is similar to the one immediately above except that once several cluster centers are chosen, the whole set of data is partitioned into disjoint sets based on nearest distance to cluster centers being within threshold distance of still other objects.

MATCHING-TYPE MEASURES

Quite often the analyst wishing to cluster profiles must contend with data that are only nominal scaled, in whole or in part. While dichotomous data, after suitable transformation, can often be expressed in terms of inter-point distances, the usual approach to nominal-scaled data uses attribute-matching coefficients. Intuitively speaking, two profiles are viewed as similar to the extent to which they share common attributes. As an illustration of this approach, consider the two profiles appearing below:

Each of the above objects is characterized by possession or non-possession of each of six attributes, where a “1" denotes possession and “0" denotes non-possession. Suppose we just count up the total number of matches –1, 1 or 0, 0– and divide by the total number of attributes. A simple matching measure could then be stated as:

S12 = M/N = 2/6 = 1/3 where M denotes the number of attributes held in common (matching 1's or 0's) and N denotes the total number of attributes. We notice that this measure varies between zero and one. If weak matches (non-possession of an attribute) are to be de-emphasized, the above measures can be modified to:

Sij=

No. of attributes that are 1 for both i and j

No. of attributes that are 1 for either i or j or both

Cluster Analysis | 7

In this case, Sij = 1/5. A variety of such matching type coefficients are described by Sneath and Sokal [1973] and Everett

[1980]. Attributes need not be limited to dichotomies, however. In the case of unordered multichotomies, matching coefficients are often developed by means similar to the above by recoding the k-state variables into k-1 dummy (zero-one) variables. Naturally such coefficients will be sensitive to variation in the number of states in each multichotomy.

Finally, mention should be made of the case in which the variables consist of mixed scales–nominal, ordinal, and interval. Interval-scaled variables may be handled in terms of similarity coefficients by the simple device of computing the range of the variable Rk and finding:

Gower (1967) suggested this measure as a means of handling both nominal- and interval-scaled data in a single similarity coefficient. Mixed scales, which include ordinal-scaled variables, present greater difficulties. If ordinal and interval scales occur, one can downgrade the interval-scaled data to ordinal scales and use nonmetric procedures. If all three scales–nominal, ordinal, and interval–appear, one is more or less forced to downgrade all data to nominal measures and use matching type coefficients.

An alternative approach would be to compute “distances” for each pair of objects according to each scale type separately, standardize the measures to zero mean and unit standard deviation, and then compute some type of weighted association measure. Such approaches are quite ad hoc, however. While the above classes of clustering algorithms are not exhaustive of the field, most of the currently available routines can be typed as falling into one (or a combination) of the above categories.

Criteria for grouping include such measures as average within-cluster distance and threshold cut-off values. The fact remains, however, that even the “optimizing” approaches generally achieve only conditional optima, since an unsettled question in this field is how many clusters to form in the first place.

The recommendation of an algorithm is difficult at best. For classification purposes, clusters should be able to identify distinct separations between different clusters of items. Clusters should also be internally consistent. Because meeting these challenges is often a function of the type of data analyzed, selection of an optimal algorithm is also a function of the characteristics of the data.

Overall, the k-means clustering technique appears to perform well (see Punj and Stewart 1983) when the initial starting configuration is non-random. In situations where a random starting configuration is required, the minimum variance type of algorithm often performs well. It is even suggested that clustering might best be approached using a combination of reduced space analysis and clustering techniques, so as to group points in the space obtained from principal components or nonmetric scaling techniques. This approach is particularly beneficial if the number of dimensions is small, allowing the researcher to augment the clustering results with visual inspection of the configuration. If the researcher is more concerned with structure than classification, overlapping clustering ignores the concept of distinct separations between clusters in an attempt to allow products/subjects to belong to more than one cluster.

Cluster Analysis | 8

DESCRIBING THE CLUSTERS

Once clusters are developed, the researcher still faces the task of describing the clusters. One frequently used measure is the centroid; that is, the average value of the objects contained in the cluster on each of the variables making up each object’s profile. If the data are interval scaled and clustering is performed in original variable space, this measure appears quite natural as a summary description.

If the space consists of principal components’ dimensions obtained from nonmetric scaling methods, the axes usually are not capable of being described simply. Often in this case the researcher will want to go back to the original variables and compute average profile measures in these terms. If matching type coefficients are used, the cluster may be described by the group’s modal profile on each of the attributes; in other cases, arithmetic averages may be computed. In addition to central tendency, the researcher may compute some measure of the cluster’s variability, e.g., average inter-point distance of all members of the cluster from their centroid or average inter-point distance between all pairs of points within the cluster. STATISTICAL SIGNIFICANCE

Despite attempts made to construct various tests of statistical significance of clusters, current statistical tests are little more than heuristics offering relatively indefensible procedures. The lack of appropriate tests stems from the difficulty of specifying realistic null hypotheses. First, it is not clear just what the universe of content is. Quite often the researcher arbitrarily selects objects and variables and is often interested in confining attention to only this sample. Second, the researcher is usually assuming that heterogeneity exists in the first place–otherwise, why bother to cluster? Moreover, the clusters are formed from the data and not on the basis of outside criteria. Thus, one would be placed in the uncomfortable statistical position of “testing” the significance between groups formed on the basis of the data itself. Third, the distributions of objects are largely unknown, and it would be dangerous to assume that they conformed to some tractable model like a multivariate normal distribution. It is indeed likely that different types of clusters may be present simultaneously in the data.

It is a major difficulty to specify a-priori the type of clustering or homogeneity to be detected. These limitations in mind,

Arnold (1979) proposed using a statistic originally suggested by Friedman and Rubin (1967). The statistic, given by

C = log

max | T |

|W|

where

| T | is the determinant of the total variance-covariance matrix

| W | is the determinant of the pooled within-groups covariance matrix

We continue to believe that, at least in the present state of cluster analysis, the objective of this class of techniques should be to formulate rather than test categorizations of data. After a classification has been developed and supported by theoretical research and subsequent reformulation of classes, other techniques like discriminant analysis might prove

Cluster Analysis | 9

useful in the assignment of new members to groups identified on grounds which are not solely restricted to the original cluster analysis.

While the above caveats are not to be taken lightly, clustering techniques are useful–in ways comparable to the objectives of factor analysis–as systematic procedures for the orderly preclassification of multivariate data. The results of using these approaches can be helpful and meaningful (after the fact) as will be illustrated next.

An Application of

Reduced Space and Cluster Analysis

Thus far, our descriptions of reduced space and clustering methods have largely remained at the conceptual level. In this section of the section we describe their application to a realistically sized problem–one dealing with the similarities and differences among 90 1987 automobiles, trucks and utility vehicles whose prices range from $5,000 to $168,000. In this abridged version of the study, we illustrate the use of cluster analysis to 90 vehicles. Table 1 identifies the 20 attributes upon which data was collected for the 90 vehicles identified in Table 2. This 90 vehicle x 20 variable data matrix forms the basis for our analysis directed at grouping the vehicles according to similarity of attributes.

Application of the Howard Harris procedure yielded two different clustering solutions which, based on the within-group sum of squares for each group, appeared to be worth examining. Figure 3 shows the “Scree”-type diagram plotting sum of squares against number of clusters. The curve appears to flatten at 5 clusters and again at 12 clusters. The 5-cluster solution is shown in Table 3 and the 12-cluster solution is shown in Table 4. As might be surmised, the 5-cluster representation was inferior to the 12-cluster representation.

Figure 3: Cluster Number by Total Sum of Squares

Cluster Analysis | 10

Table 1: Vehicle Performance Characteristics and Listing

Table 2: The 90 Vehicles Included in the Cluster Analysis

Cluster Analysis | 11

Table 3: The Five-Cluster Solution

In Table 1, we note that cluster membership is somewhat more evenly distributed than in Table 2 (twelve) clusters. The groups are rather homogeneous, though now and again, a vehicle seems to be out of place. From Tables 1 and 2 one can get some idea of the current inter-manufacturer competition. Market positioning strategies seem to be well developed, with several of the manufacturers having multiple vehicles within the same segments. This product positioning is even more apparent when one recognizes that these are the major model differences and that minor options/product distinctions are present for most vehicles.

Cluster Analysis | 12

Table 4: 12-Cluster Solution

SUMMARY OF STUDY

The foregoing results constituted only one of several possible facets of this study. Additional analytical steps may have involved: (a) the development of clusters based only on the nominal-scaled (features) data; (b) the development of clusters based only on the interval-scaled data; and (c) clustering (involving both features and measured data) on a combined time period basis.

Cluster Analysis | 13

In terms of substantive results, we found that five “clusters” explained most of the similarities and differences among the vehicles models–VW Vanagon, smaller 4-5 passenger vehicles and wagons, exotic sports and large capacity passenger cars and utility vehicles, and popular sports cars and pickups. Of course, the clusters became more detailed as the number of clusters increased.

The resulting clusters indicate which manufacturers compete with which other manufacturers in terms of similarity in the performance profiles of their vehicles. For purposes of this section, suffice it to say that clustering techniques can be used in marketing studies involving large-scale data banks. Moreover, the combination of reduced space (principal components) and cluster analysis can provide a useful dual treatment of the data. The reduced space phase may provide help in summarizing the original variables in terms of a smaller number of dimensions, e.g., speed or cargo capacity. The clustering phase permits one to group vehicles according to their coordinates in this reduced space.

Other Considerations in Clustering Techniques

Our previous discussion of clustering analysis has tended to emphasize the tandem approach of dimensional and nominal

(class-like) representation of data structures. In addition to using multidimensional scaling techniques for reduced space analysis, a number of other nonlinear approaches have been developed, including nonlinear factor analysis [McDonald,

1962], polynomial factor analysis [Carroll 1969], correspondence analysis [Carroll, Green and Schaffer, 1986]. Space does not permit anything but brief mention of this interesting work. We do consider in some detail, however, a combination qualitative-quantitative approach to an important problem in reduced space analysis–the interpretation of data structures. NOMINAL VS. DIMENSIONAL STRUCTURES

As mentioned earlier, even a pure class structure–where class membership accounts for all of the information in the data–can be represented spatially. More commonly, however, we consider cluster analysis as a more appropriate technique for characterizing such data. On the other hand, other data structures are inherently dimensional, so that measures of proximity are assumed to be able to vary rather continuously throughout the whole matrix of proximities. Pure typal and pure dimensional structures represent only two extremes. Since all proximity matrices (that obey certain properties [Gower,

1966]) can be represented spatially, it would seem of interest to consider data structures in terms of the restrictions placed on the points as they are arranged in that space. This motivation underlies many of the developments in cluster analysis. Torgerson [1965] was one of the first researchers to become interested in the problem of characterizing data as

“mixtures” of discrete class and quantitative variables. Several varieties of such structures can be obtained:

Cluster Analysis | 14

1.

Data consisting of pure and unordered class structure. Dimensional representation of such data would consist of points at the n vertices of an n-1 dimensional simplex where inter-point distances are all equal. For example, three classes could be represented by an equilateral triangle in two-space, four classes by a regular tetrahedron in three-space, and so on.

2. Data consisting of concentrated masses of points, corresponding to classes, where interclass distances are unequal, thus implying the existence of latent dimensions underlying class descriptions.

3. Data consisting of hierarchical sets of attributes where some classes are nested within other classes, e.g., cola and non-cola drinks within the diet-drink class.

4. Data consisting of dimensional variables nested within discrete classes, e.g., sweet to non-sweet cereals within the class of “processed” shape (as opposed to “natural” shape) cereals.

5. Data consisting of mixtures of ideal (mutually exclusive) classes so that one may find, for example, points in the interior of an equilateral triangle whose vertices represent three unordered classes.

6. Data consisting of pure dimensional structure in which, theoretically, all of the space can be filled up by points.

While the above categorizations are neither exclusive nor exhaustive, they are illustrative of the variety of data structures that could be obtained in the analysis of “objective” data or subjective (similarities) data of the sort described in the preceding sections. From the viewpoint of cluster analysis, some of the above structures could produce elongated, parallel clusters in which average intra-cluster distance need not be smaller than inter-cluster distances. Moreover, one could have structures in which the clusters curve or twist around one another along some manifold embedded in a higher dimensional space [Shepard and Carroll, 1966].

Figure 3: Dimensional Portrayal of Alternative Data Structures

Figure 3 shows three types of data structures as related to the above categories [Torgerson, 1965].

The first panel illustrates the case of three unordered discrete classes. The second panel illustrates the case of discrete class structure where class descriptors are assumed to be orderable. The third panel shows the case of three discrete classes and an orthogonal variable, which is quantitative. Points occur only along the solid lines of the prism. The fourth panel illustrates the case where objects are made up of mixtures of discrete classes plus an orthogonal quantitative dimension. In this case all objects lie on or within the boundaries of the curve prism while

“pure” cases would lie at one of the three edges with location dependent upon the degree of the quantitative variable which each possesses.

Cluster Analysis | 15

Research in cluster analysis and related techniques is proceeding in new directions for dealing with heretofore-intractable data structures. The continued development and refinement of interactive display devices should further these efforts by enabling the researcher to “visualize” various characteristics of the data array as a guide to the selection of appropriate grouping methods

OVERLAPPING CLUSTERING TECHNIQUES

The key element of all clustering techniques discussed so far is the mutually exclusive and exhaustive nature of the clusters developed. While in most cases, managers view segments as mutually exclusive and hierarchical in nature, cases do exist where segments are mutually exclusive. Indeed, consumers may well fit into several segments. Overlapping clustering relaxes the exclusivity constraint of most other hierarchical and non-hierarchical cluster models. As an example of a cluster analysis of brands of soft drinks, Tab may be perceived as fitting into clusters identifying diet drink, cola, and used by women, whereas Diet Pepsi would fit into only the first two benefit clusters. Brands might compete across product categories. V8 drink would compete against other vegetable/fruit drinks, as well as against soft drinks and even as a between meals snack. A cluster of toothpaste users might show that Aqua-Fresh toothpaste appeals to the fresh breath, decay prevention, and brighteners clusters, while Crest may appeal to only the decay prevention benefit cluster. Overlapping clustering simply allows for patterns of overlapping to be considered.

Arabie [1977], Shepard and Arabie [1979], Arabie and Carroll [1980], Arabie, Carroll, DeSarbo and Wind [1981] outline methods for overlapping clustering, but point out that limitations do occur in practice. First, it is difficult to develop an algorithm that effectively considers all possible cluster overlap options, especially if the sample size is large. Second, most overlapping clustering algorithms produce too many clusters with excessive overlap. A high degree of overlap results in poor configuration recovery, or in other words, a great mathematical model that is difficult to visualize from the data.

Shepard and Arabie [1979] provide a detailed explanation of their ADCLUS (for “additive clustering”) model. The ADCLUS model represents a set of m clusters which may or may not be overlapping. Each cluster is assigned a numerical weight, wk, where k=1,…,m. The similarity between any pair of points is predicted in the model as the sum of the weights of those clusters that contain the pair.

Arabie and Carroll [1980] and Arabie, Carroll, DeSarbo and Wind [1981] further develop the ability to fit the ADCLUS by presenting the MAPCLUS (for MAthematical Programming CLUStering) algorithm. This implementation appears to meet the needs of clustering items in more than a single cluster. In addition, clusters may be added, deleted, or modified to produce constrained solutions [Carroll and Arabie, 1980], and estimate (in a regression sense) the importance of new sets of clusters in explaining variance in the data.

The importance of overlapping clustering is self evident, particularly in applications where clusters are not mutually exclusive, but are overlapping. This reality reflects the existence of multi-attribute decision rules in decision-making behavior, divergent product application or use scenarios, and even joint decisions made by multiple users within the same household. Cluster Analysis | 16

Summary

This paper has considered a companion objective of the scaling of similarities and preference data–the use of metric and nonmetric approaches in data reduction and taxonomy. Clustering procedures are a helpful tool in data analysis when one desires to group objects (or variables) according to their relative similarity. We first provided a description of clustering methods and addressed the topics of association measures, grouping algorithms, cluster descriptions and statistical inference. This led to presentation of some pilot research utilizing cluster analysis, in examining the performance structure of the automobile market. We concluded the section with a description of the general problem of portraying data structures that consist of mixtures of categorical and dimensional variables and a discussion of the usefulness of overlapping clustering.

1. We note that factor analysis can be used to cluster respondents and cluster analysis can be used to group variables. This is done by transposing the data matrix (i.e., using a variable by subject data matrix rather than the more common subjects by variables data matrix).

2. Other authors provide step-by-step demonstrations of the single, complete, and average linkage rules.

We’re Here to Help!

Qualtrics.com provides the most advanced online survey building, data collection (via panels or corporate / personal contacts), real-time view of survey results, and advanced “dashboard reporting tools”.

If you are interested in learning more about how the Qualtrics professional services team can help you with a conjoint analysis research project, contact us at research@qualtrics.com.

Cluster Analysis | 17

Premium Essay

...ASB-1104 Introduction to Marketing Assignment 1 In view of the dynamic nature of the marketing environment, to what extent do you consider consumers to be, in practice, central to marketing activities? Name: ZHUOMING AN Student No: 500356688 Tutor: David James Introduction What is marketing? The answer is not changeless. There are some different definitions about marketing. The Chartered Institute of Marketing define that "Marketing is the management process responsible for identifying, anticipating and satisfying consumers' requirements profitably." (CIM). Taking a concern into this definition, it indicates that marketing begins before a product or service is developed. In additional, it also explain that marketing involves identifying an unsatisfied consumer need or want and determining if a profitable opportunity exists. Another definition is that “A social and managerial process by which individuals and groups obtain what they need and want through creating and exchanging products and value with others." (Kotler et al., 2005). The basic idea of this definition is that core to all marketing activities is customer satisfaction, which means marketing is an ongoing process as consumer demands and the environment is constantly changing. Products need to adapt as demands change. At the same time, marketing does not involve misleading, tricking or manipulation the customer. The Jobber also define the marketing is "The achievement of corporate goals through meeting...

Words: 1933 - Pages: 8

Premium Essay

...*/introduction Critically evaluates the marketing planning process Discusses impediments to effective implementation of marketing plan Introduction The leading exponents of the marketing planning have been warned of the communications factors, operational, cultural and managerial in which frequently impede the effective implementation of the marketing planning programmers in the past two decades. (Cravens, 1998; Doyle, 1998; Greenley, 1982; Leeflang and de Mortanges, 1996; McDonald, 1992a, b; 1995; Piercy and Morgan, 1994; Jain, 1993; Simkin, 1996a, b; Verhage and Waarts, 1988). There have some specific guidance are offered in the recent years to assist marketing managers overcoming those internal organisational and in pre-empting forces (cf. Cravens, 1998; Dibb et al., 1996; Lings, 1999; Piercy, 1997; 1998; Simkin, 2000). Yet, the recent research has shows barriers to the implementation of programmes and marketing strategies. (Dibb and Simkin, 2001; Simkin, 2000). Another key barrier is indicating impeding the deployment of effective marketing practices used to be the lack in most marketing function or either in organisations. (cf. McDonald, 1992a, b; Piercy and Morgan, 1994). The research are shows this is a no longer to the case with the bulkiness businesses professing to have a marketing department undertaking not only promotion and customer research,but are relate to the Kotleresque textbook approach to marketing management (Dibb and Simkin, 1997; Piercy...

Words: 1580 - Pages: 7

Premium Essay

...A high-level data model in business or for any functional area is an abstract model that documents and organizes the business data for communication between functional and technical people. It is used to show the data needed and created by business processes. A data model in software engineering is an abstract model that documents and organizes the business data for communication between team members and is used as a plan for developing applications, specifically how data are stored and accessed. An entity-relationship model (ERM) is an abstract conceptual data model (or semantic data model) used in software engineering to represent structured data. There are several notations used for ERMs. Methodology: 1. Use E-R model to get a high-level graphical view of essential components of enterprise and how they are related 2. Then convert E-R diagram to SQL DDL, or whatever database model you are using E-R Model is not SQL based. It's not limited to any particular DBMS. It is a conceptual and semantic model – captures meanings rather than an actual implementation The E-R Model: The enterprise is viewed as set of * Entities * Relationships among entities Symbols used in E-R Diagram * Entity – rectangle * Attribute – oval * Relationship – diamond * Link - line Ellipsis (plural ellipses; from the Ancient Greek: ἔλλειψις, élleipsis, "omission" or "falling short") is a series of dots that usually indicate an intentional omission of a word, sentence...

Words: 1759 - Pages: 8

Premium Essay

...Defining Marketing Paper ShuMikki Stinson MKT/351 Thomas Collins August 26, 2013 Marketing is a tool that can help business owners to prosper with their product. Marketing can be defined as an instrument that is helpful to individuals to promote their goods in their unique way that other people items would not be an eye-catcher. It is vital to know that the key thing for every marketing tool that the individual uses will be used for the proper reasons. By the individual using marketing tools in the proper way, the individual will reach their marketing level in which they wish to be at. There are side effects when it comes to the marketing tools and the techniques which they do not work properly as it was intended too. There are numerous of definitions that relate to the marketing field. According to William D. Perreault, the author of Basic Marketing, the best definition of marketing is the performance of activities that seek to accomplish an organization’s objectives by anticipating customer or client needs and directing a flow of need satisfying goods and services from producer to customer or client. (William D. Perreault, 2011). Marketing is important because it can be a benefit for a business or a hardship. Respectable marketing methods can make a transformation between a concrete upsurges in sales to an impasse circumstances on a quality merchandise. Marketing can be as simple as having a general conversation, which is usually the case in the long run. Marketing is vital...

Words: 767 - Pages: 4

Premium Essay

...Omar Rochell Marketing MKT/421 April 7, 2011 Nikki Jackson Introduction Marketing is exposed to someone every day, even when they do not seem realize it. Driving down the roads you see billboards everywhere and that is part of marketing. Logos people were on their shirts and signs in the middle or on the sign of football fields are all part of marketing. Even when a child is marketing themselves to their parents to borrow the car or go to a party they are marketing themselves to their parents in exchange for the car or the party. A set of activities that will benefit both parties’ objectives is my own personal definition of marketing. This paper will be defining marketing in different perspectives. Discussing the importance of marketing in a organizational success will also be discussed with examples included from different organizations. As an organization it is important to know what marketing is and how to establish success. What is Marketing “Marketing is defined as the activity, set of institutions, and processes for creating, communicating, delivering, and exchanging offerings that will have value for customers, clients, partners, and society at large.”(American Marketing Association, 2011) Marketing is a process that helps links the consumer, customer, and public to information that will help identify and market opportunities. Marketing research will generate, and evaluate different types of market actions, monitor marketing performance, and help improve...

Words: 1088 - Pages: 5

Premium Essay

...focus inward on the organization’s needs instead of outward (the customer’s needs). • Product is aimed at everyone. • Firms want to profit through maximizing sales volume. • Promotion to achieve goals. 2. Describe some of the characteristics of a firm that would follow a marketing orientation. Marketing orientation is “a philosophy that assumes that a sale does not depend on an aggressive sales force but rather than on a customer’s decision to purchase a product; it is synonymous with the marketing concept.” • Unlike sales orientation, a firm would focus outward on the customers wants and needs. • The goal of a firm is to satisfy customers wants and needs and delivering superior value. • The target is specific groups of people. • Where sales orientation profits by sales volume, marketing orientation firms profit with good feedback from customers or customer satisfaction. • It’s more about marketing and less about selling (less persuasion). • Firms identify what customers want and have businesses give them what they want efficiently. 3. In what ways does McDonald's embody both a marketing and a societal marketing orientation? Do some internet research if necessary. McDonald’s embodies a marketing orientation...

Words: 1110 - Pages: 5

Premium Essay

...customer-focused and heavily committed to marketing. These companies share a passion for understanding and satisfying customer needs in well-defined target markets. They motivate everyone in the organization to help build lasting customer relationships based on creating value. Marketing is just as important for non-profit-making organizations as it is for profit-making ones. It is very important to realize that at the heart of marketing is the customer. It is the management process responsible for identifying, anticipating and satisfying consumer requirements profitability. Background The term ‘‘marketing’’ is derived from the word ‘‘market’’, which refers to a group of sellers and buyers that cooperate to exchange goods and services. The modern concept of marketing evolved during and after the revolution in the 19th and 20th centuries. During that period, the proliferation of goods and services, increased worker specialization and technological advances in transportation, refrigeration and other factors that facilitate the transfer of goods over long distances resulted in the need for more advance market mechanisms and selling techniques. But it was not until the 1930s that companies began to place a greater emphasis on advertising and promoting their products and began striving to tailor their goods to specific consumer needs. By the 1950s, many larger companies were sporting entire marketing departments charged with devising and implementing marketing strategies that would complement...

Words: 2190 - Pages: 9

Premium Essay

...MARKETING PLAN RESEARCH DEFINITION: A marketing plan is a business document written for the purpose of describing the current market position of a business and its marketing strategy for the period covered by the marketing plan. Marketing plans usually have a life of from one to five years. PURPOSE: The purpose of creating a marketing plan is to clearly show what steps will be undertaken to achieve the business' marketing objectives. CONTENT OF MARKETING: A marketing plan for a small business typically includes Small Business Administration Description of competitors, including the level of demand for the product or service and the strengths and weaknesses of competitors. 1. Description of the product or service, including special features 2. Marketing budget, including the advertising and promotional plan 3. Description of the business location, including advantages and disadvantages for marketing 4. Pricing strategy 5. Market Segmentation The main contents in marketing plan are: * Executive Summary Brief statement of goals and recommendations based on hard data. * Environmental Analysis Presents data on the market, product, competition, distribution, macro-environment. (Product fact book) S.P.I.N.S. Situation “Where am I”, Problem identification/Implications “What is happening”, Needs Assessment “Why is it happening”, Solutions “What can I do about it” Market Situation: Data on target market, size and growth for past years...

Words: 579 - Pages: 3

Premium Essay

...Marketing MKT 421 Marketing According to “American Marketing Association” (2013), “Marketing is the activity, set of institutions, and processes for creating, communicating, delivering, and exchanging offerings that have value for customer, clients, partners, and society at large.” The American Marketing Society has grown to be the largest marketing associations in the world. The members work, teach, and study in the field of marketing across the globe. Another definition of marketing is according to “About.com Investors” (2013), “Marketing is an activity. Marketing activities and strategies result in making products available that satisfy customers while making profits for the companies that offer those products.” Organizations success lies in marketing and it is the heart of the success. The marketing introduces a product or service to potential customers. An organization can offer the best service or product in the industry but the potential customers would not know about it without marketing. Sales could crash and organizations may close without marketing. For a business to succeed the product or service that is provided needs to be known to the potential buyers. Getting the word out is important part of marketing in any organizational success. Product or service awareness is created by marketing strategies. If marketing is not used the potential customers will never be aware of the organizational offerings and the organization will not have the opportunity to succeed...

Words: 776 - Pages: 4

Premium Essay

...chapter 1 Marketing’s Role in the Global Economy When You Finish This Chapter, You Should 1. Know what marketing is and why you should learn about it. 2. Understand the difference between micro-marketing and macro-marketing. 3. Know why and how macromarketing systems develop. 4. Understand why marketing is crucial to economic development and our global economy. 5. Know why marketing special— ists—including middlemen and — facilitators—develop. 6. Know the marketing functions and who performs them. 7. Understand the important new terms (shown in red). www.mhhe. When it’s time to roll out of bed in the morning, does your General Electric alarm wake you with a buzzer—or by playing your favorite radio station? Is the station playing rock, classical, or country music—or perhaps a Red Cross ad asking you to contribute blood? Will you slip into your Levi’s jeans, your shirt from L. L. Bean, and your Reeboks, or does the day call for your Brooks Brothers interviewing suit? Will breakfast be Lender’s Bagels with cream cheese or Kellogg’s Frosted Flakes—made with grain from America’s heartland—or some extra large eggs and Oscar Mayer bacon cooked in a Panasonic microwave oven imported from Japan? Will you drink decaffeinated Maxwell House coffee—grown in Colombia—or some Tang instant juice? Will you eat at home or is this a day to meet a friend at the Marriott-run cafeteria—where you’ll pay someone else to serve your breakfast? After breakfast, will you head off to school...

Words: 14069 - Pages: 57

Premium Essay

...Abstract In the world of today with rude competition everywhere, customers’ expectations have become higher than ever. It is not the customers who come towards the products but it is the products which should make their way to the customers. And for this, only competitive businesses that are able to stimulate customers’ interests survive in the market. Therefore firms need to increase customers’ awareness about their products or services to be able to pull and encourage them to engage in purchase of their products. And as such, the promotional mix used by a company is really important for this task. The promotional mix in itself is very broad, consisting of various tools, like advertising, personal selling, direct marketing, public relation and sales promotion. To make the optimum use of these tools, marketers usually select them, depending on their budget and objectives, as well as the sector in which they operate (Kotler & Armstrong 1997). As such, research has been conducted on the use of promotional mix and research questions and objectives have been set. The methodology which will be used has been devised. We shall be doing a descriptive study through a survey questionnaire, in which there will be open as well as close ended questions and the questionnaire will be administered through personal interview that is direct, face-to-face. The sample size will be 100 persons and will all be customers of J Kalachand & Co Ltd. After the research, we will be...

Words: 4233 - Pages: 17

Premium Essay

...Marketing is the process of communicating the value of a product or service to customers, for the purpose of selling that product or service. Marketing can be looked at as an organizational function and a set of processes for creating, delivering and communicating value to customers, and customer relationship management that also benefits the organization. Marketing is the science of choosing target markets through market analysis and market segmentation, as well as understanding consumer behavior and providing superior customer value. From a societal point of view, marketing is the link between a society’s material requirements and its economic patterns of response. Marketing satisfies these needs and wants through exchange processes and building long term relationships. Organizations may choose to operate a business under five competing concepts: the production concept, the product concept, the selling concept, the marketing concept, and the holistic marketing concept.[1] The four components of holistic marketing are relationship marketing, internal marketing, integrated marketing, and socially responsive marketing. The set of engagements necessary for successful marketing management includes capturing marketing insights, connecting with customers, building strong brands, shaping the market offerings, delivering and communicating value, creating long-term growth, and developing marketing strategies and plans.[2] Marketing may be defined in several ways, depending on...

Words: 270 - Pages: 2

Premium Essay

...oriented philosophy is so important. The phrase market-oriented is used in marketing conversations as an adjective describing a company with a marketing orientation. Market orientation more describes the company's approach to doing business. Market-oriented defines the company itself. If a company is market-oriented, its board and executive leadership believe that the best way to succeed is to prioritize the marketplace above products. This usually goes over well with customers, but the company also must have adequate research and development to provide what the market wants. Hence, a market-oriented organization is one whose actions are consistent with the marketing concept. Difference Between Marketing Orientation & Market Oriented by Neil Kokemuller, Demand Media http://smallbusiness.chron.com/difference-between-marketing-orientation-market-oriented-14387.html Marketing is a management process and management support for marketing concept is very important element in success. If a company wants to be successful then it is market oriented. Marketing involves identifying the customer requirements and estimate the customer requirements in future. It requires planning which is very important process of marketing. To satisfy the needs the business should provide benefits – offering right marketing at right time at right place. Generally market based companies adopt strategic level marketing that defines the mission and long term objectives of the company. Market oriented...

Words: 716 - Pages: 3

Premium Essay

...qwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmrtyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmrtyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmrtyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmqwertyuiopasdfghjklzxcvbnmrtyuiopasdfghjklzxcvbnmqwer...

Words: 789 - Pages: 4

Premium Essay

...Assessment: MKC1 Market Environmental Variables Reading: Contemporary Marketing: Chapter 3 Questions: 1. How would you categorize Generation X using the five segments of the marketing environment? A: Competitive Environment B: Political-legal environment C: Economic environment D: Technological environment E: Social-cultural environment 2. Joe and Ryan both have storefronts in the local mall. Joe sells candies and Ryan sells pretzels. Are Joe and Ryan in direct competition with each other? A: Yes B: No Consumer Behavior and Marketing Reading: Contemporary Marketing: Chapter 5 Questions: 1. Rachel and Sarah’s parents always purchased groceries from the local Aldi marketplace. What is this type of behavior an example of? A: Cultural influences B: Social Influences C: Personal factors 2. Maryanne purchases Maxwell House coffee every two weeks from the grocery. What is this type of behavior an example of? A: Routinized Problem Solving B: Limited problem solving C: Extended problem solving 3. Aaron does research on several local colleges before applying to his first three choices. This is an example of: A: High – involvement purchase decision B: Low – involvement purchase decision Marketing Plans Reading: Contemporary Marketing: Chapter 2 + Ch. 2 Appendix Web sites: http://www.jpec.org/handouts/jpec33.pdf http://www.netmba.com/marketing/process/ Questions: 1. Strategies are designed to meet objectives...

Words: 8933 - Pages: 36