cutoff for # of dimensions in NMS

[gpmal...@gmail.com]

Hi All -

When running an NMS either in autopilot or with parameter setup with runs of real data >1, the output can give fewer dimensions than evaluated. For example:

Axes  Minimum     Mean  Maximum  Minimum     Mean  Maximum      p         n
---------------------------------------------------------------------------
   1   23.383   34.532   45.681
   2   11.984   12.882   13.781
   3    7.235    7.961    8.687
---------------------------------------------------------------------------
p = proportion of randomized runs with stress < or = observed stress
i.e., p  = (1 + n)/(1 + N)
      n = no. permutations <= observed
      N = no. permutations

Conclusion:  a 2-dimensional solution is recommended.

In such a case, I might want to use the third axis if I think that the change in stress is still valuable, even if <5.  I can do this for a single run of real data, but the stress and especially the per axis R2 of ordination and original distances can change quite a bit between runs (the reason for doing many and letting the algorithm choose).

Is there a way to get the third dimension or change the cutoff in change of stress?

The Help, for Ways to Run NMS, says:

4. Verify that you wish to use the number of dimensions that PC-ORD selected automatically in step 1. Beyond which additional dimensions should provide only small reductions in stress. 

But I don't see anything about what to do if you wish elsewise.

Thanks, and apologies if I am missing something obvious.

 - George Malanson

[gpmal...@gmail.com]

I think I found the answer myself - so maybe no need to post the initial question.

I unchecked the box for Step down in dimensionality.  Given that I have run the NMDS as many times as I have, I should have given that more thought.

 - George M

[Bruce McCune]

.George, in cases like this I recommend taking autopilot off and forcing the dimensionality that you want. So in this case, you know you want 3 dimensions, so after you turn off autopilot, go to Parameters tab and set number of parameters = 3. Then uncheck "step down in dimensionality", then you might as well set the "number of runs with real data" to a high number (say 200) and the maximum number of iterations to a high number (say 500).

I'm pasting below a screenshot of some "tips on controlling NMS with autopilot off" -- this is from a lab activity in my course packet.

Bruce McCune

--
You received this message because you are subscribed to the Google Groups "PC-ORD" group.
To unsubscribe from this group and stop receiving emails from it, send an email to pc-ord+un...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/pc-ord/62c889c3-b737-45f7-adad-135c81eecac5n%40googlegroups.com.