Hi,
I have a dataset with 25 outcome variables (derived from a 25-item scale), and am hoping to do some ordinal regressions with it. To test the proportional odds assumption beforehand I need to create 3 dichotomous split categories for each outcome variable, however this split format will be uniform across each of the 25 outcome items. Is there a way to do this in SPSS so I can 'bulk' apply the cumulative split format to each of the outcome variables, so I don't have to do it manually 25 times over?
I'm not the most proficient at SPSS, so would really appreciate some advice on this, thank you.
Hello Emma Young. If you use the PLUM command to estimate your model, it will give you the so-called Test of Parallel Lines in the output. That test is a test of the proportional odds assumption. Does it fail to meet your needs in some way? What do you plan to do differently with your 3 dichotomous variables?
Also, statistical tests of the assumptions of another statistical model are often not very helpful. As sample size increases, they tend to have too much power, and therefore detect departures from the assumptions that are not big enough to care about.
Finally, how many levels do your 25 items have? Your desire to convert each one to 3 dichotomies makes me think there might be 4 levels (with dichotomies 1 v 234, 12 v 34, 123 v 4), but you have not said how many levels there are.
Thanks for clarifying.
First, Bruce Weaver is a very knowledgeable SPSS Programmer and I have no doubt that his procedure will work. However 25 outcome ordinal variables seem like an awful lot to me.How many IVs do you have? Are you proposing a multivariate ordinal regression with ,all 25 DVs. ?If not how are you parsing them among the IVS? This seems a giant data set for such a statistical procedure and seems overwhelming to me. I don't do this type of survey work myself so I don't follow the details and perhaps this is standard in your field. I have never had 25 DVs in a data set before and I wonder how all of this goes and will provide interpretative results.If you would please explain and give a research question or so we could see what you are intending with all these variables.. What for example. size is your data matrix and the studied sample. If this is just my lack of experience please forgive me In any case I wish you the best in your research. David Booth
David Eugene Booth raises another very important point--i.e., the multiplicity problem when you have so many outcomes. How do you intend to deal with that?
If you need some reading material on multiplicity, the two Lancet articles by Schulz & Grimes (2005) are among the most thoughtful pieces I've come across so far.
Article Schulz K, Grimes DMultiplicity in randomised trials I: endpo...
Article Multiplicity in randomised trials II: Subgroup and interim analyses
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS