[讨论]Cluster Analysis

SUGI 23: An Interactive Graphical Interface for Hierarchical ...

Paper po26224 Lev Sverdlov The FASTCLUS Procedure as an Effective Way to Analyze Clinical DataSize: 192 Kb Keywords: FUSTCLUS procedure Cluster Analysis Clinical Data

[此贴子已经被作者于2005-12-22 12:07:50编辑过]

Paper sd26261 Bryan D. Nelson Variable Reduction for Modeling Using PROC VARCLUS Size: 142 Kb Keywords: clustering variables segmentation

Cluster analysis is an exploratory data analysis tool for solving classification problems. Its object is to sort cases (people, things, events, etc) into groups, or clusters, so that the degree of association is strong between members of the same cluster and weak between members of different clusters. Each cluster thus describes, in terms of the data collected, the class to which its members belong; and this description may be abstracted through use from the particular to the general class or type.

Cluster analysis is thus a tool of discovery. It may reveal associations and structure in data which, though not previously evident, nevertheless are sensible and useful once found. The results of cluster analysis may contribute to the definition of a formal classification scheme, such as a taxonomy for related animals, insects or plants; or suggest statistical models with which to describe populations; or indicate rules for assigning new cases to classes for identification and diagnostic purposes; or provide measures of definition, size and change in what previously were only broad concepts; or find exemplars to represent classes.

Whatever business you're in, the chances are that sooner or later you will run into a classification problem. Cluster analysis might provide the methodology to help you solve it; and Clustan could provide the professional software you need for that task.

http://www2.chass.ncsu.edu/garson/pa765/cluster.htm

Lecture Notes: Cluster Analysis

以下是引用SPSSCHEN在2005-12-22 10:54:53的发言： Hi everyone, I am looking for different ways to test the stability of clusters after they have been generated (eg using holdout cases or monte carlo simulations etc). I appreciate that the type of test will vary across clustering alogorithms but I would appreciate some generic reference material (preferably internet accessible) for different ways to assess cluster stability and listed criterion that could be applied to the assessment process. Regards Paul

One approach when using a single method is the so-called "split sample" method. Steps are: - Divide the sample into two, and perform a cluster analysis on one of the samples, having a fixed rule for the number of clusters. - Determine the centroids of the clusters, and compute proximities between the objects in the second sample and the clusters, classifying the objects into their nearest cluster. - Cluster the second sample using the same methods as before, and compare these two alternative clusterings for the second sample. References for the split sample method include: McIntyre, R.M. and Blashfield, R.K. (1980), A nearest-centroid technique for evaluating the minimum variance clustering procedure. Multivariate Behavioral Research, 22, 225-238. Breckenridge, J.N. (1989), Replicating cluster analysis: Method, consistency and validity. Multivariate Behavioral Research, 24, 147-161. Both of these papers are referenced in Cluster Analysis, 4th edition, by Everitt, Landau, and Leese. When comparing two alternative clusterings, you can use the adjusted Rand index. The adjusted Rand index was introduced by Hubert and Arabie: Hubert, L.J., and Arabie, P. (1985), Comparing partitions. Journal of Classification, 2, 193-218. When using different methods, you can synthesize the results using *consensus clustering*. Cheng and Milligan have written about assessing the influence of individual points. Cheng, R. and Milligan, G.W. (1996), Measuring the influence of individual data points in a cluster analysis. Journal of Classification, 13, 315-335. Anthony Babinec

Dear All,

Can any one please tell me how to compute Item Discrimination? That is point-biserial item-total correlation for each item in a test.

Thanks Humphrey

以下是引用hanszhu在2005-12-23 10:38:30的发言：

Dear All,

Can any one please tell me how to compute Item Discrimination? That is point-biserial item-total correlation for each item in a test.

Thanks Humphrey

Point-biserial correlation is simply the correlation between a dichotomous item and the total score (continuous/scale) that may or may not contain the item. Zachary zfeinstein@harrisinteractive.com

Does anyone know where I can find a Macro in SPSS to run HC robust standard errors when my data is clustered?  Basically, I have a sample where some of my observations are in clustered, and need to calculate "real" standard errors and probabilities.  I have a Macro for the standard HCREG correction, the test developed I believe by White 1980.

Thanks

it's too complicated