+ Reply to Thread
Results 1 to 7 of 7

Thread: Repeated measures, random effects, or both?

  1. #1
    Points: 3,653, Level: 38
    Level completed: 2%, Points required for next Level: 147

    Posts
    107
    Thanks
    9
    Thanked 0 Times in 0 Posts

    Repeated measures, random effects, or both?




    Hi, I have the following situation and I do not know if it is a case that calls for repeated measures or individual-specific random effects.

    We have subjects that are followed for a fixed amount of time. If an event happens in the time window then we treat it as an event, if an event does not happen by the end of the fixed time then it will never happen and treat it as a nonevent (rather than a censored observation). Much like a mortgage - if you default before paying it all off, then it's an event. If you don't default by the end of the term, then you won't ever default.

    Subjects are followed across time and the response is recorded. The data is then transposed from wide to long format. I modeled on the long data using standard logistic regression, and compared the results versus modeling it using GEE, treating subjects as the repeated measures. I tried various pre-defined covariance structures and none performed as well as the standard logistic regression. The caveat is that there is no variable selection in GEE – I just take the same variables as the ones in the final logistic model and fit in in the GEE. Should the subjects be adjusted for as repeated measures or random effects?

    Please for the sake of this issue, assume that fitting a time series or a completely new model is not an option.

  2. #2
    Omega Contributor
    Points: 38,432, Level: 100
    Level completed: 0%, Points required for next Level: 0
    hlsmith's Avatar
    Location
    Not Ames, IA
    Posts
    7,006
    Thanks
    398
    Thanked 1,186 Times in 1,147 Posts

    Re: Repeated measures, random effects, or both?

    I usually think about repeated measures (outcome) as random effects. They both control for observations within a cluster.
    Stop cowardice, ban guns!

  3. #3
    Points: 3,653, Level: 38
    Level completed: 2%, Points required for next Level: 147

    Posts
    107
    Thanks
    9
    Thanked 0 Times in 0 Posts

    Re: Repeated measures, random effects, or both?

    Sorry I don't understand... To put it in SAS language that I do understand...

    repeated=id

    or

    random=id?

    or

    random=y (binary event indicator)?

  4. #4
    Omega Contributor
    Points: 38,432, Level: 100
    Level completed: 0%, Points required for next Level: 0
    hlsmith's Avatar
    Location
    Not Ames, IA
    Posts
    7,006
    Thanks
    398
    Thanked 1,186 Times in 1,147 Posts

    Re: Repeated measures, random effects, or both?

    I listened to a lecture from a SAS Corp person a couple of weeks ago and they strongly encouraged using the "repeat". Tell us more about these data (what is repeated, is outcome binary or continuous, etc.).


    perhaps you are looking at:


    repeat int / subject=id;
    Stop cowardice, ban guns!

  5. #5
    Points: 3,653, Level: 38
    Level completed: 2%, Points required for next Level: 147

    Posts
    107
    Thanks
    9
    Thanked 0 Times in 0 Posts

    Re: Repeated measures, random effects, or both?

    What's the difference between repeated int / subject=id vs repeated subject=id?

  6. #6
    Omega Contributor
    Points: 38,432, Level: 100
    Level completed: 0%, Points required for next Level: 0
    hlsmith's Avatar
    Location
    Not Ames, IA
    Posts
    7,006
    Thanks
    398
    Thanked 1,186 Times in 1,147 Posts

    Re: Repeated measures, random effects, or both?

    Not at a computer, so I can upload a picture, but pretty much each person can have a different slope and intercept. A quick Internet search of images should help. So you would be saying that they don't just vary by angle but also where they intersect the intercept.
    Stop cowardice, ban guns!

  7. #7
    Points: 3,653, Level: 38
    Level completed: 2%, Points required for next Level: 147

    Posts
    107
    Thanks
    9
    Thanked 0 Times in 0 Posts

    Re: Repeated measures, random effects, or both?


    Thank you, this next question is specific to SAS: I am fitting a GEE with an unstructured covariance matrix. How can I output the parameter estimates and use those to calculate estimates on another dataset?

+ Reply to Thread

           




Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts






Advertise on Talk Stats