+ Reply to Thread
Results 1 to 6 of 6

Thread: Do I need to control for Type 1 errors?

  1. #1
    Points: 150, Level: 2
    Level completed: 50%, Points required for next Level: 50

    Posts
    16
    Thanks
    0
    Thanked 0 Times in 0 Posts

    Do I need to control for Type 1 errors?




    Hi guys, the situation I have is, a reviewer has asked me to reduce my mediation model (MPLUS) as he says I have too many tests going on. In fairness it is a large model.

    its 6 IVs > 4 mediators > 1 DV so I guess 24 possible mediated tests although I only have 3 significant indirect paths. i have complied and reduced the model to 5 iVs 3 mediators but I dont wish to lose anymore variables as this will upset the focus of the paper.

    I guess I could use Bonferroni corrected alpha but i dont know how to calculate corrected 95%Ci - so that is out.

    So at the moment - I have 15 IVs on 3 mediators and 1 DV. Personally, I think this is okat - but what do you think?

  2. #2
    Omega Contributor
    Points: 38,303, Level: 100
    Level completed: 0%, Points required for next Level: 0
    hlsmith's Avatar
    Location
    Not Ames, IA
    Posts
    6,993
    Thanks
    397
    Thanked 1,185 Times in 1,146 Posts

    Re: Do I need to control for Type 1 errors?

    What is your sample size? Are any of the IVs categorical with greater than 2 groups?
    Stop cowardice, ban guns!

  3. #3
    Points: 150, Level: 2
    Level completed: 50%, Points required for next Level: 50

    Posts
    16
    Thanks
    0
    Thanked 0 Times in 0 Posts

    Re: Do I need to control for Type 1 errors?

    Hi,

    my sample size is 7000 people and all variables are binary (0,1).

  4. #4
    Omega Contributor
    Points: 38,303, Level: 100
    Level completed: 0%, Points required for next Level: 0
    hlsmith's Avatar
    Location
    Not Ames, IA
    Posts
    6,993
    Thanks
    397
    Thanked 1,185 Times in 1,146 Posts

    Re: Do I need to control for Type 1 errors?

    By mediation regression, you are alludung to having multiple interaction terms? Are all of your terms significant or hold clinical significance?
    Stop cowardice, ban guns!

  5. #5
    Points: 150, Level: 2
    Level completed: 50%, Points required for next Level: 50

    Posts
    16
    Thanks
    0
    Thanked 0 Times in 0 Posts

    Re: Do I need to control for Type 1 errors?

    Quote Originally Posted by hlsmith View Post
    By mediation regression, you are alludung to having multiple interaction terms? Are all of your terms significant or hold clinical significance?
    out of 15 tests, 3 indirect effects are significant and they hold theoretical value more than applied clinical significance. Path model below. Neglect and Recreation has now been removed.

    https://www.dropbox.com/s/a425kwomm5...model.PNG?dl=0

  6. #6
    Points: 140, Level: 2
    Level completed: 80%, Points required for next Level: 10
    seanw's Avatar
    Location
    Bloomington, IN
    Posts
    22
    Thanks
    0
    Thanked 6 Times in 6 Posts

    Re: Do I need to control for Type 1 errors?


    Quote Originally Posted by DrFurbs View Post
    Hi guys, the situation I have is, a reviewer has asked me to reduce my mediation model (MPLUS) as he says I have too many tests going on. In fairness it is a large model.

    its 6 IVs > 4 mediators > 1 DV so I guess 24 possible mediated tests although I only have 3 significant indirect paths. i have complied and reduced the model to 5 iVs 3 mediators but I dont wish to lose anymore variables as this will upset the focus of the paper.

    I guess I could use Bonferroni corrected alpha but i dont know how to calculate corrected 95%Ci - so that is out.

    So at the moment - I have 15 IVs on 3 mediators and 1 DV. Personally, I think this is okat - but what do you think?
    It's not too difficult to calculate the CIs if you use Bonferroni correction. Basically, when you change the p-value, you adjust the critical t-value accordingly when you calculate your CI boundaries. Let p* be your Bonferroni-adjusted p-value and df be your degrees of freedom (same as it would be); then, you can find the corresponding critical t-value in R with

    qt(1-p*,df)

    Substitute the new critical t-value into the CI formula and voila.

    Conversely, if it's not strictly necessary to report CIs, you could simply report p-values and compare them to the Bonferroni-adjusted p-value. If you report a standard error as well, then knowledgeable statisticians can figure out the adjusted CIs themselves.
    - Sean W.
    http://www.helpmewithstats.com/

+ 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