Hi there!
I would like to perform a generalized linear mixed effects model but am not sure about the best method for my datastructure. My design is as follows:
- There was a total of 24 vignettes that belong to 3 need clusters (8 in each cluster).
- Each participant read 12 comparisons of 2 vignettes, each from a different cluster. Each vignette contained information about one person with one need and a gender.
- Thus, the comparisons always consisted of 2 different needs, each in combination with a different gender
I would like to find out which variables influence the choice of a vignette and include in my model:
- The need of the vignettes (3 need-clusters)
- The content of the vignettes (24 different contents)
- Gender in the vignettes
- Nationality of participants (sample consists of only two nationalities)
- potential interactions
- Random intercept
I would like to perform a generalized linear mixed effects model (in SPSS) but am unsure about how to work with the fact that participants always saw only 2 vignettes at a time in each comparison. If I used “cluster chosen” as dependent variable I would not take into account that participants never chose between all three clusters at a time. They read the same amount of vignettes out of each cluster but the comparison was always just between two.
I have information about
o Vignette_Person1: need, gender, content of vignette
o Vignette_Person2: need, gender, content of vignette
o Participant: nationality
o Cluster chosen
o Gender chosen
o What kind of Comparison (need1 vs. need2; need 1 vs. need 3; need 2 vs. need 3)
I also got the warning “Hessian Matrix is not positive definite although all convergence criteria are satisfied. Validity of model fit is uncertain.” several times even with just one predictor variable. Do you have an idea how to avoid this warning?
I would be very thankful for advise! Thank you very much in advance!
I think the warning about the Hessian matrix not being positive definite is usually related to model specification or quasi-complete separation in the data. This often happens in logistic models when one predictor (or combination of predictors) almost perfectly predicts the outcome. A few things you could try:
The warning does not always mean the results are invalid, but it signals that the standard errors or fit indices may be unreliable. Carefully check model convergence, parameter estimates, and whether the predictors are really supported by the sample size
Your data must have been measured at least twice over time. Then a simple search for "GLMM model in R" will bring up a simple example.
- Badges
- Science topic
- Similar topics
- Phytochemistry
- Extracts