Hello -
My unit has recently upgraded to SPSS v16. We conduct a large number of analyses using Correspondence Analysis. Those familiar with CA may recall that when you receive new data, one can run and obtain new points for the new data that will be comparable to the original data. The way this is done is by (1) merging the original and new data, (2) specifying the new data should be treated as "supplemental" to the analysis, and (3) re-running the analysis with the same options and configuration as before.
In previous versions, the results from a CA with supplemental points was very reliable/repeatable - meaning that for the original data, we ALWAYS got identical results for the data that was part of the original study. We never observed the problem I am about to describe.
The Problem
CA provides dimension scores as part of the output. These scores, in 2 or more dimensions, can be used to create a map/chart of the results. As I mentioned, previous versions always provided consistent results and dimension scores for the original data, no matter if supplemental data were used or not.
In SPSS v16, it seems that while the general interpretability of the results is consistent, the dimension scores for an analysis involving supplemental data are different from the original. Specifically, the new provided coordinates for original items has either reverse signs in one or both coordinates. For these supplemental analyses, these coordinates are the only piece we are interested in.
I can provide some output if needed to further explain.
The issue is that we generate these supplemental maps often enough to have put in place SPSS production job code that automates our process. The fact that the coordinates can come opposite to the original creates massive investigation challenges.
Questions
This issue would call is either a "Feature" or a "bug" of v16 Categories module. The mathematics of CA are such that this problem shouldn't occur. Does anyone else have experience with this problem w- v16?
Also, I have contacted SPSS and have yet to hear from their support on this. For anyone who has also experienced this challenge, is there a possible workaround to either check and correct (automatically) or avoid the problem in the first case?
I would be glad to provide more information if necessary, but wanted to stick to the main points in the initial post. Thanks.
Brandon L. Paris
Manager, Analytic Innovation & Methods
Center for Learning & Experimentation/Consumer Insights
General Mills, Inc.
====================To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD
Locked
