Erik-Andr� SAULEAU wrote: > Hi, > > I have individuals related by a similarity (real between 0 and 1, which is > not a distance) and I want to clusterise individuals. Actually I have a > symetric matrix like this: > subject 1 2 3 ... > 1 1 .2 .7 ... > 2 .2 1 .8 ... > 3 .7 .8 1 ... > ... > > In fact it is possible that exist some incoherence (a very similar to b, b > very similar to c but a not similar to a). Perhaps have I two questions? > 1-how to transform my similarity (a complex index) in distance for using in > a classical clusters analysis? > 2-how to use directly my similarity (like a kind of discriminant analysis > for correlation between variable)? > > Thanks in advance for your help. > Erik A. > > -- > ============================================ > Erik-Andr� SAULEAU > > SEAIM > H�pital du Hasenrain > BP 1070 > 87, Avenue de Altkirch > 68051 MULHOUSE C�dex > > Tel: 03-89-64-79-95 > Fax: 03-89-64-79-71 > M�l: [EMAIL PROTECTED] > Web: http://www.ch-mulhouse.fr > ============================================ > > Any of the standard clustering procedures can use a correlation matrix. I distribute a free PC version of the Lance-Williams-Boesch combinatorial agglomerative clustering procedure called COMPAH. Jardine & Sibson (1968) argued that only single-linkage clustering was compatiable with a correlation matrix, but most would argue (including Lance & Williams) that the other clustering methods can be used with correlations. The documentation for the program has more details.
. . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
