Hi all, I am trying to run moderated mediation analysis using Process Macro model n°8 with 8 mediator variables but I got this error message: "ERROR: A linear or near linear dependency (singularity) exists in the data." mainly in the part of Y results.
Does anyone working with Process Macro know how to solve this error ?
PS: I m working with SPSS 25 version
Hi Olfar,
this error can occur for example if you recorded your variable but there was a mistake and now you effectively have a DV without variation (e.g., the DV was 0/1 and you decided it and now it is 1 for all overs actions). Another reason might be that you have split the data file or told SPSS at some point to analyze group wise (See ”split file“). At least that has happened to me before. Finally, this can happen if some of the variables are coded as string or numerical with string. PROCESS cannot deal with this and you have to transform them.
Hope it helps!
Sure thing, Olfa Ammar. Did you end up solving the issue? Or did you find out if anything else was happening? To reiterate my previous response: Most times, this error signals that there is not sufficient variation in an independent variable and (in an experimental design) there is no between groups variance to explain deviations in the dependent variable.
Hi Alexander Mafael , actually that message only appear when I analysed the model with all the mediator variables (8 mediators), but if I analyse the model with one mediator variable at a time everything works well, and this is actually what I did.. but I don't know how correct is it..
Dear Olfa Ammar I think it is best if you just use smaller models anyways - otherwise interpreting the different effects gets a little messy. Also, do you have specific hypotheses regarding a parallel or sequential mediation with eight different mediators at the same time? My best guess would be: No. At least I have never seen a paper (not even with field data) use eight mediators simultaneously. I would stick to your hypotheses and analyze the mediation effects accordingly. For example, if you believe that M1 and M2 simultaneously mediate an effect, focus on this model first. Next, report the results for (as an example) the model where M2 and M4 sequentially mediate an effect, and so on.
I hope that I was clear: In my humble opinion you should only actually test for mediation effects where you have a concise hypothesis or an underlying logic as to why you believe these relationships exist. Simply running a model with a ton of mediators will be a hard sell to reviewers anyway, but I also do not see the added value in doing so. It just complicates matters. In addition, you probably end up having a problem justifying the causality that is implied in your design.
Let me know if you have further questions.
Best,
Alexander
Yes I totally agree with you, it's gonne be very complicated and hard to discuss and explicate the results.
Thank you very much Alexander Mafael .
best,
Hi Alexander Mafael , maybe you could help me with a similar question. So I received the same error message as Olfa, while I was running my moderated-mediation analysis. My IV and moderator are both categorical. (IV coded: 1,2,3 - moderator coded: 0,1). I assume that the problem lies in my moderator as I could perform my process mediation with no problem. Is there any advice you could give me? Many thanks in advance
Hi Melanie Ernstbrunner, I read in another thread that apparently this error should not occur when you are using SPSS Version 22. Two other things that can be a problem is potential multicollinearity between your IV and the moderator or a case where certain values of a moderator only exist for certain values of the IV (e.g., the moderator "0" only exists in IV "1" - probably very unlikely). In that case, there is no variation between the groups that can be used to explain variance in the DV. Let me know if you find the reason!
I had the error when something was apparently not set to numeric. Do as.numeric() on all the variables you wish to PROCESS.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Structural Equation Modeling