Discussion: View Thread

  • 1.  Power Calculation for mixed effect multiple logistic regression models

    Posted 03-02-2012 17:38
    Dear All,

    I would really appreciate if some one can provide me references for power calculations in case of multiple logistic regression and/or mixed effect  multiple logistic regression where predictors are categorical as well as continuous random variable. My sample size is 566 and there are about 12 predictors.

    Also, if you can point towards some references for the procedures available in SAS for such power estimation.

    Thank you and I look forward for the groups suggestions.

    Have a nice weekend.
    -------------------------------------------
    [Tasneem] [Zaihra]
    [Assistant Professor]
    [Concordia University]
    [Montreal]
    [QC]
    [Canada]
    -------------------------------------------


  • 2.  RE:Power Calculation for mixed effect multiple logistic regression models

    Posted 03-02-2012 17:45
    The SAS PROC POWER procedure has a worked example for binary logistic regression
    with multiple predictors.

    In my opinion, the problem with power calcs for complex models is not so much the mechanics
    of the calculation -- it's the complexity of having the investigator provide the appropriate inputs
    to the calculation.  You can always try to run a simulation in the absence of other calc tools. 

    Ever try to get an investigator to give you the presumed correlation between pre and post
    measurements in a simple paired t-test calculation? 

    Hope this helps.

    Marty

    -------------------------------------------
    Martin Lesser, PhD
    Director, Biostatistics Unit
    Feinstein Institute for Medical Research
    Manhasset, NY 11030
    -------------------------------------------








  • 3.  RE:Power Calculation for mixed effect multiple logistic regression models

    Posted 03-02-2012 19:09

    I believe that PASS (excellent power analysis software) does this. In the logistic regression setup you specify the R-squared for the independent variable of interest linearly regressed on the other independent variables. I think this is nice, because you don't have to specify anything variable by variable--just the R-squared. 

    It isn't SAS, but perhaps you will have access to PASS, or some reader will come up with something simple for SAS. 

    Best wishes,

    Nayak


    -------------------------------------------
    Nayak Polissar
    Consultant
    The Mountain Whisper Light
    -------------------------------------------