+ Reply to Thread
Results 1 to 7 of 7

Thread: Any stats tests I can run to show inconsistency/ no correlation?

  1. #1
    Points: 18, Level: 1
    Level completed: 35%, Points required for next Level: 32

    Posts
    4
    Thanks
    0
    Thanked 0 Times in 0 Posts

    Any stats tests I can run to show inconsistency/ no correlation?




    Hi, I am doing my msc dissertation and I have gathered 204 responses to a questionnaire with one of the main questions being what exercises specific professionals give at certain stages post op. My hypothesis is that there is inconsistency between what exercises specific professionals give during rehab.
    Basically I have four stages of rehab - early, intermediate , late and return to sport and I have 16 exercises and lots of numbers about how many are given at each stage.
    However as there are zero and less than five in quite a few parts chi squared won't work or the df comes out at 45!

    I also have four groups on different levels of experience and whether otherwise I could order the exs in most to least given for each stage and maybe compare experience to say there is lack of conformity?!

    Or am open to any other assistance/ comments, have trawled through to textbooks, all I can find online and our uni tutors are being really unhelpful.

    Many thanks

  2. #2
    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: Any stats tests I can run to show inconsistency/ no correlation?

    The common solution when you have less than 5 cases in cells is to collapse the level of one or more of the variables being analyzed. I think Fischer's exact text might also apply - you might look that up. If one of your variables has 16 levels chi square probably is not the best test. I have rarely seen it used with more than 5 levels. Personally it seems like you should collapse the 16 exercises into fewer levels for example on something that makes theoretical sense (this exercise is similar to this one and different from that group based on some theory) or else consider factor analysis to support collapsing them.

    It is not clear to me if you are testing that the same individuals give different exercises at each level of the rehab or individuals give different exercises at each level (from other individuals).
    "Very few theories have been abandoned because they were found to be invalid on the basis of empirical evidence...." Spanos, 1995

  3. #3
    Points: 18, Level: 1
    Level completed: 35%, Points required for next Level: 32

    Posts
    4
    Thanks
    0
    Thanked 0 Times in 0 Posts

    Re: Any stats tests I can run to show inconsistency/ no correlation?

    Hi thanks for this.....yes they are the same sixteen exercises throughout, I tried to combine them but still some exercises were zero at some stages! May be attempting the impossible I feel?!

    I wondered otherwise about breaking them down to %s but still numbers are that different that they don't work!

    If anyone has any other suggestions would be appreciated , or feel just some nice graphs showing distributions are coming on!

  4. #4
    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: Any stats tests I can run to show inconsistency/ no correlation?

    I think that you probably will need to get them down to 4 or 5 levels to work in chi square. If individuals are consistantly assigning the same exercise across rehab than the proportion of each exercise should remain constant over level of rehab . So if the proportion changes this is a way of testing that they are not assigning the same exercises (or more accurately that as a group they are not assigning the same exercises) across levels of the rehab. You might be able to use a test of proportions. Or at least show the descriptive change in proportion.
    "Very few theories have been abandoned because they were found to be invalid on the basis of empirical evidence...." Spanos, 1995

  5. #5
    Points: 18, Level: 1
    Level completed: 35%, Points required for next Level: 32

    Posts
    4
    Thanks
    0
    Thanked 0 Times in 0 Posts

    Re: Any stats tests I can run to show inconsistency/ no correlation?

    Hi Noetsi, really appreciate your tips, far more helpful than our stats people! Am thinking you mean combine exercises so only five/ six variations then re run . Sorry to ask further advice, when you mention a test of proportion, what sort of thing do you mean?

    Many thanks, really helpful

    Gards79

  6. #6
    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: Any stats tests I can run to show inconsistency/ no correlation?

    Well for starters I suggest talking over my suggestions with other stats people you know - because I have been known to be wrong And if your committe does not agree with the approach even if the approach is formally right it does not matter. Your committee is never wrong...

    thinking you mean combine exercises so only five/ six variations then re run
    yes. If I understand correctly you are doing a chi square test with one of your variables being rehab levels (4) and the other exercise which has 16 distinct levels. 16 is as you found out, simply too many distinct levels in practice to use chi square with. So I am reducing these levels to some manageable set say 4 or 5. One way to do this is collapse exercise down by putting similar exercises (or ones seen as such in the literature) into the same category. The other is to use factor analysis to collapse them.
    "Very few theories have been abandoned because they were found to be invalid on the basis of empirical evidence...." Spanos, 1995

  7. #7
    Points: 18, Level: 1
    Level completed: 35%, Points required for next Level: 32

    Posts
    4
    Thanks
    0
    Thanked 0 Times in 0 Posts

    Re: Any stats tests I can run to show inconsistency/ no correlation?


    Thanks again, will attempt to regroup exercises on similar but can I ask what you mean by factor analysis to collapse them?

    Many thanks

+ 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