HELP! Questionnaires and implants -Which statistical test? Multi Level Models?

Home Forums Methodspace discussion HELP! Questionnaires and implants -Which statistical test? Multi Level Models?

Viewing 5 posts - 1 through 5 (of 5 total)
  • Author
  • #2232
    Mort Bones

    I am in a complete pickle and need some help.

    I am interested in whether a particular operation (implant) improves sleep quality and whether the type of implant matters.


    I have two patient groups.

    One had implant A and the other implant B. 

    I measured the patients quality of sleep before surgery, and then at 1month, 6 months and 12months post- operation

    To measure sleep quality I used a questionnaire (PSQI).  The PSQI consists of 9 questions, which are scored on a 3-point Likert scale) The 7 components of sleep are derived from the answers and then these are added up to give a Global Score. 

    I also collected data about their gender, co-morbidities, 1st or 2nd operation of this type, and age.


    1. What is the difference between pre-op (baseline) and 1month, 6months and 12 months post-op for each implant type and across implant types.

    2. what is the contribution of the other factors – age, gender, co-morbuidities etc to the results.


    I also have some missing data which is common with patient data.  There are 300 complete data sets where the patients completed the questionnaire and all 4 time points and there are about 300 that did not.


    I am using SPSS and am familiar with it for doing chi square, binomials, one way ANOVA, Wilcoxens but for this set of data I have been working through linear mixed models but I am not sure whether this is the right test?!  I am in a right pickle. Please Help.

    Rafael Garcia

    Given the knowledge constraints, I would do a mixed GLM. The difference scores (baseline – 1month; baseline – 6 months; baseline – 12months) as your outcomes and time (the within measure), age, sex, and co-morbidities as the predictors. The output will give you the Between effects of most of the predictors and the Within effect of time.

    There seem to be step by step instructions of something similar here:

    Hope that helps,


    Mort Bones

    That is really helpful. Thank you!  What does the Covariates option mean in there? If I find an effect of gender and then want to know if there is an effect of age but control for gender do I then put gender in the Covariates and age in the between-subjects factor bit?

    Thank you SO MUCH!

    Rafael Garcia

    I don’t advocate using ‘covariate’ options. What is a covariate really? Something that you measured that effects the Outcome that you say a priori, “I’m not interested in that.”

    If you have a reason to think something affects your Outcome Variable, add it as a predictor. Better yet, add it hierarchically (SS1). That will control for the early predictors and tell you what whether the later variables are significant after controlling for that.

    As an aside, I found another resource here:



    Muir Houston

    According to the version of the PSQI on their website, it does not have a three point likert scale – it is a four point ordinal scale.

    i attach it here

    and are you using their scoring database?

    have you checked here:

    and why do you not follow the analysis of the designers of the test? If you read this paper:

Viewing 5 posts - 1 through 5 (of 5 total)
  • You must be logged in to reply to this topic.