+ Reply to Thread
Results 1 to 10 of 10

Thread: Ordinal regression vs Chi-Square

  1. #1
    Points: 2,514, Level: 30
    Level completed: 43%, Points required for next Level: 86

    Posts
    24
    Thanks
    0
    Thanked 0 Times in 0 Posts

    Ordinal regression vs Chi-Square




    Hey guys,
    I've finished analyzing my data for an article regarding decompresion in divers.
    My dependent variable is ordinal - 1-no recovery 2-partial 3-compelte recovery, and the independant variable is the time it took to treat the patients.
    When i do an ordinal logistic regression - i get p=0.048.
    However when i thought about showing the assosciation - i tried to divide the time to brackets such as 48-71 hours, 72-96 hours >96 hours and i've done CHi-square for the table of 3X3 - which turns out insignificant.

    So what's the real thing? is there association? if so how would i show it?

    thanks for your help
    Amir

  2. #2
    TS Contributor
    Points: 17,749, Level: 84
    Level completed: 80%, Points required for next Level: 101
    Karabiner's Avatar
    Location
    FC Schalke 04, Germany
    Posts
    2,540
    Thanks
    56
    Thanked 640 Times in 602 Posts

    Re: Ordinal regression vs Chi-Square

    is there association? if so how would i show it?
    Use the Spearman rank correlation.

    With kind regards

    K.

  3. #3
    Points: 2,514, Level: 30
    Level completed: 43%, Points required for next Level: 86

    Posts
    24
    Thanks
    0
    Thanked 0 Times in 0 Posts

    Re: Ordinal regression vs Chi-Square

    Can u please explain?
    Y correlation here instead.of.regression?

  4. #4
    Points: 2,514, Level: 30
    Level completed: 43%, Points required for next Level: 86

    Posts
    24
    Thanks
    0
    Thanked 0 Times in 0 Posts

    Re: Ordinal regression vs Chi-Square

    also - this will try to correlate an ordinal variable with a numeric continuous one. i thought you can't do it. Also dont' you lose the ordinal importance?

    i ran it and came out non significiant.

  5. #5
    Fortran must die
    Points: 58,790, Level: 100
    Level completed: 0%, Points required for next Level: 0
    noetsi's Avatar
    Posts
    6,532
    Thanks
    692
    Thanked 915 Times in 874 Posts

    Re: Ordinal regression vs Chi-Square

    Power might be the issue, why one test is showing signficant and another not. Generally you should not take an interval value like time and convert it into an ordinal one. You lose information (a concept I don't really understand, but I know is strongly frowned on by statisticians who do).

    While spearman is commonly used with ordinal variables polychoric/tetronic is also useful. Some recommend it over spearman if you can make the, unknowable, assumption that there is a latent continuous variable behind the measured ordinal ones. Many theories of logistic regression make this assumption anyhow so it may make more sense to use this with your data.

    http://en.wikipedia.org/wiki/Polychoric_correlation

    One down side is that this is not simple to do in most software. SAS uses a special macro and SPSS (amusingly) uses R.

    I have not heard that you can not use Spearman or polychoric correlations with one interval variable. You can certainly do Chi Square tests of independence with these (that is one of each).
    "Very few theories have been abandoned because they were found to be invalid on the basis of empirical evidence...." Spanos, 1995

  6. #6
    Points: 2,514, Level: 30
    Level completed: 43%, Points required for next Level: 86

    Posts
    24
    Thanks
    0
    Thanked 0 Times in 0 Posts

    Re: Ordinal regression vs Chi-Square

    Quote Originally Posted by noetsi View Post
    Power might be the issue, why one test is showing signficant and another not. Generally you should not take an interval value like time and convert it into an ordinal one. You lose information (a concept I don't really understand, but I know is strongly frowned on by statisticians who do).

    While spearman is commonly used with ordinal variables polychoric/tetronic is also useful. Some recommend it over spearman if you can make the, unknowable, assumption that there is a latent continuous variable behind the measured ordinal ones. Many theories of logistic regression make this assumption anyhow so it may make more sense to use this with your data.

    http://en.wikipedia.org/wiki/Polychoric_correlation

    One down side is that this is not simple to do in most software. SAS uses a special macro and SPSS (amusingly) uses R.

    I have not heard that you can not use Spearman or polychoric correlations with one interval variable. You can certainly do Chi Square tests of independence with these (that is one of each).
    thanks. not gonna mess around with that. not familiar with that one.
    so i got a n-s spearman, s regression, and n-s chi sq. if i go with the regression i have no way to show the data visually.

  7. #7
    Fortran must die
    Points: 58,790, Level: 100
    Level completed: 0%, Points required for next Level: 0
    noetsi's Avatar
    Posts
    6,532
    Thanks
    692
    Thanked 915 Times in 874 Posts

    Re: Ordinal regression vs Chi-Square

    Have you tested the power of the methods you are using? My guess is that each has different power in this case and that is why you are comming up with some signficant and some not.

    I am not sure what you mean by not showing the data visually, but I would imagine there is some table or graphical approach that can show nearly anything in regression. Also purely descriptive results don't have to come from regression even if you are using regression to test the hypothesis.
    "Very few theories have been abandoned because they were found to be invalid on the basis of empirical evidence...." Spanos, 1995

  8. #8
    TS Contributor
    Points: 17,749, Level: 84
    Level completed: 80%, Points required for next Level: 101
    Karabiner's Avatar
    Location
    FC Schalke 04, Germany
    Posts
    2,540
    Thanks
    56
    Thanked 640 Times in 602 Posts

    Re: Ordinal regression vs Chi-Square

    Quote Originally Posted by XPeriment View Post
    also - this will try to correlate an ordinal variable with a numeric continuous one. i thought you can't do it. Also dont' you lose the ordinal importance?

    i ran it and came out non significiant.
    I cannot assess whether the ordinal regression analysis met all
    assumptions.

    What you mean by non-significant, of course I do not know. Could be
    p=0.050 or p=0.98.

    Regarding usage and interpretation of Spearman's rank correlation,
    there's tons of material in the net.

    With kind regards

    K.

  9. #9
    Points: 2,514, Level: 30
    Level completed: 43%, Points required for next Level: 86

    Posts
    24
    Thanks
    0
    Thanked 0 Times in 0 Posts

    Re: Ordinal regression vs Chi-Square

    Karabiner -
    Spearman's p=0.707.
    Still I though regression was the way to go inssead

  10. #10
    Points: 2,514, Level: 30
    Level completed: 43%, Points required for next Level: 86

    Posts
    24
    Thanks
    0
    Thanked 0 Times in 0 Posts

    Re: Ordinal regression vs Chi-Square


    Motesi- all assumptions are ok for my regression. I am not sure how i check the power for each test

    and still - where lies the truth then?

    about visualizing - i wanted to visualize the significant regression in graph...and figured i can't do it without bracketing the independent variable (time) to brackets...

+ 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