SPSS Statistics

Expand all | Collapse all

Error messages SPSS 28.0 Generalized linear mixed models

  • 1.  Error messages SPSS 28.0 Generalized linear mixed models

    Posted Tue July 06, 2021 01:25 PM
    I am getting the following error messages using GENLINMIX:

    1)Vorhergesagte Wahrscheinlichkeiten können nicht gespeichert werden, wenn das Ziel ein stetiges Messniveau aufweist.

    2) glmm: Die endgültige Hesse-Matrix ist nicht positiv definit, obwohl alle Konvergenzkriterien erfüllt wurden. Die Prozedur wird trotz dieser Warnung fortgesetzt. Die nachfolgend erstellten Ergebnisse beruhen auf der letzten Iteration. Die Gültigkeit der Anpassungsgüte des Modells ist ungewiss.

    3) Datenstruktur: Es wurde mindestens ein Subjektfeld angegeben, nicht jedoch tatsächlich in der Analyse verwendet.

    What does this mean? How can I avoid this messages?

    I am getting all results, but how can I interpret them?

    Maike Bestehorn

  • 2.  RE: Error messages SPSS 28.0 Generalized linear mixed models

    Posted Wed August 04, 2021 12:51 PM

    Perhaps it would help the Community to know better what these warnings were:

    1) Predicted probabilities cannot be stored if the target has a steady measurement level.

    2) glmm: The final Hesse matrix is not positively defined, although all the convergence criteria have been met. The procedure continues despite this warning. The results that are created below are based on the last iteration. The validity of the adaptation quality of the model is uncertain.

    3) Data structure: At least one subject field was specified, but not actually used in the analysis.

    David Dwyer

  • 3.  RE: Error messages SPSS 28.0 Generalized linear mixed models

    Posted Mon August 16, 2021 09:49 AM
    Hi Maike,

    It's impossible to give detailed instructions on something like this given only this information.

    The first message seems to bbe indicating that the dependent variable is defined as having scale or "continuous" measurement level, so predicted probabilities cannot be saved.

    The second message about the Hessian matrix not being positive definite is common when one tries to fit a model with too many effects given the observed data. I'd suggest trying to fit the fixed-effects model with just a residual error to start, and see if that works without the error, then add any random or repeated effects one at a time from simplest to most complicated in order to see how far it will go without the error.

    I assume the last message is because all cases from at least one subject have missing data.

    David Nichols