SPSS Statistics

multilevel logistic regression

  • 1.  multilevel logistic regression

    Posted 16 days ago
    Edited by Jessica Long 16 days ago
    Hi all, I need some help here.
    Here is the multilevel logiesstic regression i run in spss:


    GENLINMIXED
    /DATA_STRUCTURE SUBJECTS=Group
    /FIELDS TARGET= satisfaction_binary TRIALS=NONE OFFSET=NONE
    /TARGET_OPTIONS DISTRIBUTION=BINOMIAL LINK=LOGIT REFERENCE=0
    /FIXED EFFECTS= AgeGroup Sex EthnicGroup Q1...USE_INTERCEPT=TRUE
    /RANDOM USE_INTERCEPT=TRUE SUBJECTS=Group COVARIANCE_TYPE=VARIANCE_COMPONENTS
    /BUILD_OPTIONS TARGET_CATEGORY_ORDER=ASCENDING INPUTS_CATEGORY_ORDER=ASCENDING MAX_ITERATIONS=100 CONFIDENCE_LEVEL=95 DF_METHOD=RESIDUAL COVB=MODEL PCONVERGE=0.000001(ABSOLUTE) SCORING=0
    /EMMEANS TABLES = AgeGroup COMPARE = Agegroup CONTRAST = DEVIATION
    /EMMEANS TABLES = Sex COMPARE = Sex CONTRAST = DEVIATION
    ....
    /EMMEANS_OPTIONS SCALE=ORIGINAL PADJUST = SEQBONFERRONI.

    Base size is around 10,000
    Classification
    Overall Percent Correct = 65.7%

    The problems I notice is EMMEAN the standard error is big, lower and upper confidence level ranges from nearlly 0 to 1.
    This is obviously not useful. The adj sig values are 1 from deviation constasts table.
    I also notice the grand mean is very low .104, in the data the mean of satisfaction_binary it is around 0.53.

    My first question is why the grand mean is so low? I guess the multilevel plays a role  but not sure how. Is there any docuemention to explain the grand mean?

    Second question is what is causing the emmean have 0 to 1 for lower/upper confidence level? I have checked the correction between the fix effects, i do not see anything alarming. i also run the correction between satisfaction_binary and fixed effects variables, i could not find anything either.

    Another strange thing is, when i remove one of the fixed effect varaibles e.g. Q2, it lessen the above issue and the grand mean is "improving" to be higher (0.46). Since I do not know how grand mean is derived, it is hard for me to understand why one variable has so much impact. Q2 seems not highly correlated with all other variabless that involved.

    In what sitution it makes emmeans not work? What situation would make 1 varaible makes a big change on the grand mean and up/lower confidence range for the emmeans?

    I could not find any indicator to suggest any issues with the model and i know the emmeans provide odd resutls.

    Any comments or suggestion would be helpful.

    ------------------------------
    xx xx
    ------------------------------