Dear Sama,

Please find these attachments. In work sheet, sheet no. 2 is used for analysis of cluster. If there is any problem in understanding for these cluster analysis, you are free to ask.

I think this is helpful for you.

Best regards from INDIA

Find another file having cluster analysis as sample vise.

I have attached both files, one have (GraphAppPrintJob.pdf) cluster analysis as parameter vise and other (GraphAppPrintJob2.pdf) as sample vise.

Which one is applicable used.

You can use STATISTICA. The option you want is available in that which is easy to use unlike SPSS.

And I guess there is a confusion about the scale. The scale is always same. Its upto you to use it or not to use it. If you wanted to add scale, D(link)/(Dmax) * 100, just take it to coral draw or within excel only you can add this function.

Like you may be knowing Dlink/Dmax represents linkage distance. And it is linkage distances for a particular case divided by maximum linkage distance. Finally the quotient is multiplied by 100 to standardize the linkage distance (Y-axis).

I never use to provide scale in my studies. In cluster analysis, method used is more important; whether similarity or dissimilarity proximity procedure is used and with each of them which analysis is performed thereafter.

Dear Sama,

Please find herewith attached graph and the transposed excel file data.

it is better to used Weka or RapidMiner. my best Recommendation is using R tool ! you are welcome if you will have any query

regards

Muddsair

thank you Mr. Dharmendra, the plots are interesting... what program you used to generate the cluster?

Dear Dr. Mukesh, is there a way to download STATISTICA for free? how can i add scale D(link)/(Dmax) * 100 to the excel function?

Dear Dr. Javaid, your graph is interesting and thank you for indicating the scale (Dlin/Dmax*100) however, the x scale did not indicate the samples names! i would like to know what software you used to generate the cluster?

Dear Mr. Muddsair, thank you so much. i don't know how to use R and i tried to download RapidMiner but i could not get it done!

Dear Sama,

I am using GCD kit 2.3 software for cluster and other statistical programs. This is free available program.

Check out R hclust (http://stat.ethz.ch/R-manual/R-patched/library/stats/html/hclust.html) and R cutree (http://stat.ethz.ch/R-manual/R-patched/library/stats/html/cutree.html)

Try to use Statistica. It has option to make the scale D(link)/(Dmax) * 100.

I took your data and ran a simple hclust (see cluster1.jpeg) with the following lines on R:

x

Oops. Ran the wrong analysis. Should be using:

y

thank you Adam impressive work. i appreciate your contribution

Dears, to those who is following this question and their answers: you can see that three or four of the contributors to this question have develop cluster analysis and provide me with their dendogram result (thanks for there effort). each one has compleatly different classification tree! however, they used the same data set! which one should i follow?

Dear Sama,

(Sorry for the big dig up... I hope my answer would still help you).

When I choose to compute dendongram, I firstly compute a PCA.

Why firstly compute a PCA : because PCA will explain bisection organization of the dendongram. And then I could be able to criticize the dendongram results.

On the PCA results (attach figure - I standardize your data because of the lack of unit homogeneity in your variables) you see that there is 3 main group of individual.

Group A) The ones on the right

Group B) The ones on the left

Group C) The only one on the left down

This should appears on dendongram.

Then question is :

What should be the first bisections of the dendongram : It should correspond to the most explain principal component (i.e. the first which explain 61.70% of your data set variance --> horizontal difference), then others bisections should correspond to vertical difference.

Therefore we should find : (A) be on one side and (B) + (C) be on the other side.

So, I compute a dendogram (attach figure) and found that all of my hypothesis are right. I assume that I can trust that dendongram.

Finally for sure there is no one Dendogram (there is several script to compute them - here I used Ward method) as there is not only one PCA graph. You could also represent your subject with the principal component one and three, if one main variable is not well represented by the 1st or 2nd principal component and depending on the eigenvalue/inertia of the third principal component.

However all your variables are well represented in principal component 1 and 2... (see final attached figure) so I think that close this question. :-)

Good Luck,

Cyril

KIndly use STATISTICA for analysis, it provides the option for Dlink/Dmax