I am testing a number of potential indirect effects between a predictor and an outcome variable utilising the Process Macro for multiple mediators with model 4. I have six mediators, one independent variable, one dependent variable and two covarietes.
Several error messages appear in the output:
>Error encountered in source line # 7143
>Error # 12417
>Source operand is singular for INV.
>Execution of this command stops.
Can anyone know what the problems might be? I desperately need a bit of help.
Many thanks in advance,
I don't know for certain, but do any of your variables have very large values, such as salary, as these can sometimes cause problems. If so just divide by 100, 1000 or some other number to make the values manageable.
Hi, I have the same problem, also after rescaling the variable that causes the problem. I have this problem with model 10, by the way. But like in Duygu's case, it is the mediator that is causing the problem. I'm on SPSS V26 and PROCESS V3.5. Can anyone help?
Some errors are explained here for folks who come across this and had similar issues: http://processmacro.org/faq.html
Answer: This error is usually generated in response to a singularity in your model. A singularity occurs when one variable on the right hand side of a model equation is a perfect linear combination of the other variables on the right hand side of that equation. When this happens, a variable's tolerance is zero and computation cannot proceed. This can occur when you give data to PROCESS with this singularity, or when a singularity is produced during the bootstrapping procedure. One common source of this error is when one of your variables on the right hand side of the equation is dichotomous (or one of several indicator or dummy variables in the model) with relatively few cases in one of the categories. Although there is code in PROCESS to try to stop the computations from occurring in the presence of a singularity, it won't always detect one every time that it occurs.
Try running your PROCESS command again after first setting the number of bootstrap samples to zero. If the error goes away, then the singularity is being generated during the bootstrapping routine. The only solution to this is to eliminate or greatly reduce the chance of a singularity occurring by, for example, collapsing categories of a multicategorical variable when one of the categories has very few cases. If you are using a model that allows for the construction of Monte Carlo confidence intervals, choose the Monte Carlo option instead of bootstrapping for inference about indirect effects.
When a singularity occurs, this will often produce additional error messages in the output, such as an "out of range" error. This is a consequence of the singularity, not a new error. @Duygu Uzun
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Structural Equation Modeling