3
$\begingroup$

A colleague and I have been clustering some data in SPSS (v19) and R (2.15), respectively. Using the same distance metric and agglomeration method, we get identical merge orders/agglomeration schedules in both programs, and the dendrograms have very similar shapes, but the actual height values are quite different. On this particular data set, the R dendrogram is about 150 units tall, but the SPSS dendrogram is only 25 units tall. This is somewhat $\ldots$ unsettling, obviously.

While thumbing through some manuscripts and webpages, I noticed that all but one of the SPSS-derived dendrograms were also exactly 25 units tall (example 1, 2, 3; we also have a pretty hefty pile of papers that are unfortunately all pay-walled, including the single counter-example).

The caption on the SPSS output says something about rescaling, but the documentation is oddly silent about if, how, and why SPSS might be rescaling the dendrograms.

Could someone please confirm that SPSS does rescales dendrograms (and rescales them onto [0,25])? For extra credit, is there a way to turn this rescaling off? It seems to cause SPSS to cut our dendrogram a few levels above the leaves.

Improve this question
$\endgroup$
4
  • 3
    $\begingroup$ SPSS always rescales it, step1= 1 step_last= 25. You can't change these units to original colligation values taken from agglomeration history, as far as I know. Of course, you could do it manually in any graphical editor. $\endgroup$
    – ttnphns
    Commented Dec 3, 2012 at 5:15
  • $\begingroup$ Thanks--is there any reason it does that? It's something of an anti-feature. $\endgroup$
    – Matt Krause
    Commented Dec 3, 2012 at 6:13
  • 1
    $\begingroup$ Well, I guess it is because until recently SPSS printed dendrograms in text format, so it was more handy to display same labels (1 thru 25) whatever real coefficients were. They then probably decided to let it be. It is inconvenience, but minor, for me. $\endgroup$
    – ttnphns
    Commented Dec 3, 2012 at 8:30
  • $\begingroup$ Fair enough, I guess. In a semi-related note, thanks for the collection of SPSS macros on your webpage! $\endgroup$
    – Matt Krause
    Commented Dec 3, 2012 at 20:09

1 Answer 1

Reset to default
3
$\begingroup$

Judging by examples on the web, it does rescale them.

I'm not sure changing the scaling would help with your problem of it cutting levels though.

Edit / addition (drawing on research from the OP)

The documentation here states (clicking on the dash-underlined "dendrogram"):

"The dendrogram rescales the actual distances to numbers between 0 and 25, preserving the ratio of the distances between steps."

Improve this answer
$\endgroup$
1
  • 1
    $\begingroup$ It looks like the leaves end up at zero. I finally (and accidentally) found confirmation of this in the docs: publib.boulder.ibm.com/infocenter/spssstat/v20r0m0/… and click on the dash-underlined "dendrogram. If you want to edit that into your answer, we'd be all set here! $\endgroup$
    – Matt Krause
    Commented Dec 3, 2012 at 20:15

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.