0
$\begingroup$

I'm trying to run a generalized linear mixed model in SPSS, with a continuous response variable (ResponseTime), SubjectID as a random factor, Items as a repeated measure, and five fixed factors: three categorical subject variables (Sex, Handedness, Familial Handedness) and two categorical item variables (Prime Type, Probe Latency), and one continuous covariate (Probe Size-not a variable of interest). I think the problem I'm having may have to do with specifying the random effects structure and/or data structure, but after reading everything I can find on it, nothing has helped and I'm at my wit's end. The main error message I get is this one:

glmm: The maximum number of iterations was reached but convergence was not achieved. Output for the last iteration is displayed. The procedure continues despite this warning. Subsequent results produced are based on the last iteration. Validity of the model fit is uncertain.

I am also sometimes getting this error:

glmm: The final Hessian matrix is not positive definite although all convergence criteria are satisfied. The procedure continues despite this warning. Subsequent results produced are based on the last iteration. Validity of the model fit is uncertain.

I have tried increasing the number of iterations, but to no avail; it just seems to exacerbate another problem I'm having, which is that SPSS seems to run forever (sometimes hours) when I use the repeated measures command. This seems to be a problem with the software (https://www.ibm.com/support/pages/running-repeated-measures-generalized-linear-mixed-models-and-it-just-hangs-and-runs-days-and-have-force-quit-application-mac). I don't mind waiting hours for the model to run, but I'd like to solve the nonconvergence error problem first. ANY ADVICE WOULD BE MUCH APPRECIATED!

Currently I have the following syntax:

GENLINMIXED /DATA_STRUCTURE SUBJECTS=SubjID REPEATED_MEASURES=Item /FIELDS TARGET=ResponseTime TRIALS=NONE OFFSET=NONE /TARGET_OPTIONS DISTRIBUTION=INVERSE_GAUSSIAN LINK=IDENTITY /FIXED EFFECTS=Sex Handedness FamilialHandedness PrimeType ProbeLatency Probesize PrimeTypeProbeLatency FamilialHandednessECType FamilialHandednessProbeLatency FamilialHandednessPrimeTypeProbeLatency HandednessPrimeType HandednessProbeLatency HandednessPrimeType*ProbeLatency USE_INTERCEPT=TRUE /RANDOM USE_INTERCEPT=TRUE COVARIANCE_TYPE=VARIANCE_COMPONENTS SOLUTION=FALSE /BUILD_OPTIONS TARGET_CATEGORY_ORDER=ASCENDING INPUTS_CATEGORY_ORDER=ASCENDING MAX_ITERATIONS=1000 CONFIDENCE_LEVEL=95 DF_METHOD=RESIDUAL COVB=MODEL PCONVERGE=0.000001(ABSOLUTE) SCORING=0 SINGULAR=0.000000000001

Improve this question
$\endgroup$
3
  • $\begingroup$ For some reason the asterisks in the interactions didn't come through in my post, but they are in fact there in the syntax. $\endgroup$
    – Doug
    May 17, 2020 at 23:02
  • $\begingroup$ If spss runs forever, I would say try R $\endgroup$
    – kjetil b halvorsen ♦
    May 19, 2020 at 4:24
  • $\begingroup$ You left out the most important piece of data: How many subjects are there and what is the average number of observations per subject? Also what have you done to see that the compound symmetric correlation structure dictated by this model matches the observed correlation structure, e.g., do you draw a semi-variogram? $\endgroup$
    – Frank Harrell
    Oct 1, 2021 at 12:24

1 Answer 1

Reset to default
0
$\begingroup$

It's difficult to say anything with certainty, but I would suggest removing the RANDOM subcommand and see if things converge. If this is a standard repeated measures analysis, you don't need that along with the REPEATED.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.