+ Reply to Thread
Results 1 to 4 of 4

Thread: Help with Repeated Measures Data Analysis

  1. #1
    Points: 367, Level: 7
    Level completed: 34%, Points required for next Level: 33

    Posts
    16
    Thanks
    1
    Thanked 0 Times in 0 Posts

    Help with Repeated Measures Data Analysis




    Hello,

    I have a very limited set of repeated measures data that are assessment scores over three years, with 1,055 participants who have a range of 2 to 11 completed health assessments.

    I have multiple independent variables, including demographics such as age group, gender, and race/ethnicity categories (NOMINAL). My dependent variables are mean dimension scores and composite scores, both are SCALE/CONTINUOUS variables.

    I am having issues with controlling for intervention time and drop-outs in my analysis.

    I have 1,055 first and last assessments. And I created Six-Month, 1-Year, 18-Month, and 2-Year variables if assessments occurred within the following date ranges after the first assessment: 5-7 months, 11-13 months, 17-19 months, 23-25 months.

    I tried both GLM Repeated Measures ANOVAs and Paired t-test (for pairs of first and last and each of the created time periods). I think I have too few data points for GLM Repeated Measures ANOVAs. Besides, I cannot account for seasonality.

    When I used paired t-tests, I tried them with the method of eliminating missing variables listwise and by analysis. When I did this listwise, the N became so small that I had to drop the 18 month follow-up period, and even then, I had only an N=16.

    Is this the best I can do or what do you recommend as an alternate approach for this analysis?

    Thank you,

  2. #2
    Omega Contributor
    Points: 38,334, Level: 100
    Level completed: 0%, Points required for next Level: 0
    hlsmith's Avatar
    Location
    Not Ames, IA
    Posts
    6,998
    Thanks
    398
    Thanked 1,185 Times in 1,146 Posts

    Re: Help with Repeated Measures Data Analysis

    You don't want to use repeated ANOVA it drops missing data. You want to use mixed model, which wont drop missing measures (and you can imput other missing data if desired) and you can use a variance structure that includes time between measurements. You should create a variable that is exact time between measurements.

    Is there an actual seasonality effect you are trying to measure?
    Stop cowardice, ban guns!

  3. The Following User Says Thank You to hlsmith For This Useful Post:

    AnneS (04-14-2016)

  4. #3
    Points: 367, Level: 7
    Level completed: 34%, Points required for next Level: 33

    Posts
    16
    Thanks
    1
    Thanked 0 Times in 0 Posts

    Re: Help with Repeated Measures Data Analysis

    Thank you, hlsmith.

    The subjects in this quality-improvement project analysis received health assessments at vastly different intervals. I guess the seasonality effects are less important a consideration than just the overall drop-outs. Drop-outs are inherent because this is not really a study protocol, but more of a clinical guideline that encouraged clinicians to do these assessments twice a year. But that also depends on patients attending regular visits. And there is a great deal of movement in and out of the system - attending community-based partner organizations, etc.

    Although I am concerned with "intervention" time, I really only mean that I hope to only compare those with similar exposure time, that is, time getting treatment in the health system. And the N for each time goes down if I limit it to exact months, e.g. 6, 12, 18 and 24.


    For the GLM, I did not need the created follow-up time variables (I only used those for paired t-tests). SO, I used a GLM mixed model (with a between-subjects factor = gender). And this is more of an SPSS question, but I am not sure of the Ns... where do I find them in the output? Can I deduce them from the DF?

    I see when I include a grouping factor (Gender), for example with 3 composite factors, it shows a table for Between-Subjects Factors with N for each - in this case, 283 females and 330 males. I assume that means the analysis is therefore reduced from 1,055 people to 613. And from there, it is of course, limited to those with three sequential assessments.

    I got a Box's M of 5.45, F=.903, p=0.492. Because not significant, does this mean the null hypothesis holds, i.e., that we can assume covariance is equal for the within subjects design?

    Then for within subjects effects: And I got Wilk's Lambda of F=4.572 and sig of .011 and Mauchley's W of .945, sig of .000 and Epsilons of Greenhouse-Geisser = .948 and Huynh-Feldt of .953, so I assume I use the Hyunh-Feldt correction for sphericity violation, which produces a result of F=5.65 (1.97, 1168.99), p=0.004 for time alone (but not significant with covariance with gender).

    So if all these means there are significant differences in these assessment results over time, I want to know how many people or assessments are included in that result.

    Thanks for any advice!

  5. #4
    Points: 367, Level: 7
    Level completed: 34%, Points required for next Level: 33

    Posts
    16
    Thanks
    1
    Thanked 0 Times in 0 Posts

    Re: Help with Repeated Measures Data Analysis


    Wow, I realized that MIXED MODEL you suggested may not be what I thought it was. So I'm experimenting with LINEAR MIXED MODEL using Generalized Linear type.

    I have to figure out how to run this - to set this up correctly.

    All I know is I need to go back to my long data set before restructuring.

    And read up on how to do this as I think I goofed it the first time - as it took forever to run and then gave me a bunch of information such as that the sample is normally distributed- but nothing else of interest.

+ 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