Data behind the Hierarchical Cluster Analysis

[Ayaik]

Dear Professor,
I am using Kh Koder version 3.Alpha.15f on Windows 10, I want to have the data behind the hierarchical cluster analysis, for example, the number of clusters I excluded are 9 clusters by Ward's method and Euclid distance. When I export document - code matrix and use the same method in SPSS, the 9 clusters solution shows different results of combined elements. so how can I get the data like (attached) the detailed Agglomeration schedule table and the proximity matrix in KH CODER ? to have all the data set behind the final results. appreciating your advice professor.
KH coder results.pdf
spps results.pdf
spps table.pdf

[Ayaik]

What options rather than ward's method and Euclid distance shall I choose to have the same exact results as the KH CODER clusters results because the clustering in Kh coder makes more sense with my data

[ko-ichi-h]

Can I see the cluster analysis option screen of KH Coder?

[Ayaik]

[ko-ichi-h]

Hmm, what happens if you select "Squared Euclidean distance" in SPSS? Can you get the same results as KH Coder?

[Ayaik]

the same happens if I use "Squared Euclidean distance" the 9 clusters are different than what is exported from Kh coder.

[ko-ichi-h]

Can you specify which parts are different?

I see both of them have similar clusters. For example, both of them have a cluster containing:

• TheSOHO
• TradepiaOdaiba
• DaibaFrontier
• TokyoWangan
• BMWGroup
• Time24Building
• TIEC

The orders in the dendrograms are somewhat different but members are the same, aren't they?

Also, the second clusters in the both results contain same codes:

• GrandNIKKO
• HiltonOdaiba
• FujiTvWanganStudio
• BayShuttleBus

Again, the order is different but members (codes) are the same.

[Ayaik]

yes the first four clusters are the same, but if we cut at the same distance, all the following clusters are scattered into more clusters

[ko-ichi-h]

the first four clusters are the same

The first SEVEN clusters are the same, I think. The only difference is that "Yumenohashi Bridge", "Round1" and "Sony Museum" belong to cluster (8') instead of (9').

So, does selecting "Squared Euclidean distance" in SPSS change this part?

[ko-ichi-h]

Anyway, you can export data behind the analysis in this way:

1. Save the dendrogram as a "R Source" file in KH Coder
2. Start R that came with KH Coder and run the "R Source" file saved
3. Type write.csv(d, "c:/khcoder3/cls_data.csv") in the R console and hit "enter" key to export standardized data matrix
4. Type write.csv(dj, "c:/khcoder3/cls_dist.csv") in the R console and hit "enter" key to export distance matrix
5. Type hcl$merge and hit "enter" key, then type hcl$height and hit "enter" key to view the agglomeration process. Type help(hclust) and hit "enter" to see info on how to read this.

About step 1 and 2, see page 3-6 of this slides:
https://www.slideshare.net/khcoder/example-of-using-r-1-exporting-the-result-of-correspondence-analysis

[Ayaik]

the problem occurs in cluster 7 and 6 as shown in the picture, ok I will try to export the data, thank you really professor, you are so kind and helpful

[ko-ichi-h]

Is there any problem in cluster 7 and 8? Would you explain what the problem is?

Clusters are identical between KH Coder and SPSS (squared euclid dist), aren't they?

[Ayaik]

if we cut at the same distance we will get 3 clusters or a big cluster combining 7 and 8

[ko-ichi-h]

I think the problem will be solved if you just specify "9" as the number of clusters in SPSS.

SPSS's "Rescaled" dendrogram is made with characters like "-", "+" or "L". Thus it can not describe small differences. So, it only seems to be that way. There is a small difference in the distances, I think.

[Ayaik]

I am really thankful for your help professor, you are so supportive with immediate responses which takes a lot of your time. THANK YOU!

[ko-ichi-h]

No problem at all. Good luck with your research.

BTW, this is how SPSS "rescale"s the dendrogram.

[Efaq]

Hello!

I have been struggling to find the rule that SPSS is using to "squash" the dendogram, as you showed above. Do you know how it is being calculated?

[ko-ichi-h]

Hello,

Sorry, I am not sure. I just think the "Rescaled" dendrogram is not so nice...

[Efaq]

Thank you for the response!

I completely agree. My struggle is because I need to replicate this visualization, and, although the scaling 1-25 is clear, the way they project the approximate levels is mysterious to me.

Thank you anyways!