Uniform ordination axes scales
Home › Forums › PATN and Pattern Analysis › PATN Discussion › Uniform ordination axes scales
- This topic is empty.
- November 2, 2005 at 5:18 am #415rainmanMember
Is it possible to have the same scale (min and max) on all ordination axes so that the distances between points can be related between axes?
Tom.November 8, 2005 at 7:23 am #480leeKeymaster
Basically ‘no’. As mentioned in the post to Rainman, SSH in PATN is a multidimensional scaling algorithm and the axes are somewhat arbitrary. Take a look at the Help on SSH and see how the algorithm works. It is very simple (and very effective).
Basically the scaling is standardized (one unit of distance) from an originally random set of coordinates in the selected dimensionality. The relationship BETWEEN the points in the cordinate system is not arbitrary.
I’m not sure exacly what you mean by “distance between points can be related between axes?”.
You could calculate the distance between points by exporting the ordination and then importing it as data and calculating the Euclidean Distance between objects.
LeeNovember 25, 2005 at 12:26 am #487rainmanMember
Thanks for that Lee.
While I understand that the relationship between the points in the plot is the most important feature, as you explain above, I am still a little confused.
Why are the scales on the axes different, but the actual length of the axes the same on the screen? Doesn’t this essentially squash the plot?
For example, in the following analysis I ordinated the [mean abundance of] aquatic invertebrate communities at four sites over five sampling rounds (i.e. Site 1, Round 2 = 12 etc…).
This is the patn plot. Note the nice-ish division between the dry season samples on the left (numbers ending in 1 or 4), and the wet season samples on the right (numbers ending in 2, 3 or 5).
The x axis covers a range of 1.695, while the y axis covers a range of 2.128. However, the actual axes themselves are the same length.
Exported into sigmaplot, the ordination looks like this (note the even axes). The relationship between the points (which you stress above) has changed because the axes are no longer squashed.
Is this a bug in patn, or am I missing the ‘point’ 😉 (i.e. are the axes actually measuring different things (therefore actually different scales)?
Or, is it because as you state here (http://patn.org/phpBB2/viewtopic.php?t=34), the “result will not have axes that in any way correspond to the SSH axes. SSH simply aims to get the maximum information in the selected number of axes.”
Here is the dataset f.y.i. (the Unidentified variable was made extrinsic, and the columns were log transformed prior to analysis, Bray Curtis w/ default settings).
And the project background:
Tom (user id = rainman)November 27, 2005 at 6:16 am #488leeKeymaster
You are right. It would appear that the scaling of the SSH axes in the 3d-plot are not correct. They have been scaled unit length. I had not picked up on this before as I had assumed they were scaled according to the axis length.
We have now fixed this for 3.04. We have also (due to user requests) increased the scaling difference on the front-back axis to further emphasize the 3d nature of the balls.
Just goes to show you that it is hard to detect everything in PATN – so thanks for picking this one up!
- You must be logged in to reply to this topic.