V gone -ve definite for block 0 iteration aborted (random pass, offsets 1)
Posted: Tue Mar 19, 2019 4:22 pm
Dear community,
I ran into trouble trying to model the following model in mlwin (version 3.03):
In the model we would like to estimate chances of persisting in an educational programme. So, our dependent variable is chance of persistence, which is measured as a continuous variable (0 - 10). However, the data collection was designed as a factorial survey, that is, we used vignettes to measure students' chances of persistence under variable circumstances. More specifically, students were asked to rate 6 vignettes whichvaried along two dimensions (one dimension being study success [categorical] and a second dimension being study progress requirements [yes/no]). About 500 students rated all vignettes, meaning we have a sample of about 2900 observations (of persistence chances).
The study is mutilevel by design, that is to say, we assume that the variance in persistence chances is partially on the level of the vignettes (Level 1) and partially on the level of the student (Level 2).
We already ran into problems, while defining the 0-model only taking into account the first level.
The problem was the following; V gone -ve definite for block 0 iteration aborted (random pass, offsets 1), which resulted in a 'numerical error calculating likelihood' (see attachments).
Can anyone advise us on the underlying problem and/or how to go about this?
Many thanks,
H.
I ran into trouble trying to model the following model in mlwin (version 3.03):
In the model we would like to estimate chances of persisting in an educational programme. So, our dependent variable is chance of persistence, which is measured as a continuous variable (0 - 10). However, the data collection was designed as a factorial survey, that is, we used vignettes to measure students' chances of persistence under variable circumstances. More specifically, students were asked to rate 6 vignettes whichvaried along two dimensions (one dimension being study success [categorical] and a second dimension being study progress requirements [yes/no]). About 500 students rated all vignettes, meaning we have a sample of about 2900 observations (of persistence chances).
The study is mutilevel by design, that is to say, we assume that the variance in persistence chances is partially on the level of the vignettes (Level 1) and partially on the level of the student (Level 2).
We already ran into problems, while defining the 0-model only taking into account the first level.
The problem was the following; V gone -ve definite for block 0 iteration aborted (random pass, offsets 1), which resulted in a 'numerical error calculating likelihood' (see attachments).
Can anyone advise us on the underlying problem and/or how to go about this?
Many thanks,
H.