+ Reply to Thread
Results 1 to 10 of 10

Thread: Negative effect size?

  1. #1
    Points: 920, Level: 16
    Level completed: 20%, Points required for next Level: 80

    Posts
    36
    Thanks
    16
    Thanked 0 Times in 0 Posts

    Negative effect size?




    Hi,

    I calculated the effect size for the results of Mann-Whitney U tests but some of them are negative. I was double checking the formula and how to do it and I noticed that some of them reported a positive effect size despite the formula (by doing it myself) gave a negative effect size.

    So, my question is, when you do the formula for the effect size, should you use a negatize/positive Z or should I just use the positive value?

    The formula I found is Z/sq of N.

    Thank you,

    Statn00b

  2. #2
    R must die
    Points: 25,023, Level: 95
    Level completed: 68%, Points required for next Level: 327
    noetsi's Avatar
    Posts
    4,536
    Thanks
    278
    Thanked 730 Times in 700 Posts

    Re: Negative effect size?

    You can't have a negative effect size, it is a physical impossibility. It would mean that there was less than no difference between groups which can not happen. One group could be less or higher than the other and depending on which group you make the reference group the sign of the effect might show as plus or minus - but the effect size is the absolute value of what you find. In other words the effect is the difference between one group and the other.

    If A has a mean of 5 and B of 2 then you would get a value of 3 or -3 depending on which you subtracted from which. But the effect is 3.
    This was not what we did in logistic regression. Rather, we transformed the conditional expected value, and made that a linear function of X. This seems odd, because it is odd..

  3. #3
    ggplot2orBust
    Points: 35,306, Level: 100
    Level completed: 0%, Points required for next Level: 0
    Awards:
    User with most referrers
    trinker's Avatar
    Location
    Buffalo, NY
    Posts
    3,949
    Thanks
    1,356
    Thanked 754 Times in 675 Posts

    Re: Negative effect size?

    Dason I disagree but maybe I don't understand.

    Cohen's d is an effect size and you can have a negative effect. If product B is worse than product A it will have a negative effect on the outcome of Y in comparison to A.
    "If you torture the data long enough it will eventually confess."
    -Ronald Harry Coase -

  4. #4
    TS Contributor
    Points: 5,883, Level: 49
    Level completed: 67%, Points required for next Level: 67
    jpkelley's Avatar
    Location
    Vancouver, BC, Canada
    Posts
    440
    Thanks
    17
    Thanked 90 Times in 84 Posts

    Re: Negative effect size?

    As trinker says, a negative effect is entirely possible. For the Mann-Whitney effect size, you typically take the absolute value of the formula you mentioned (like noetsi mentioned). That simply gives you the magnitude of the effect, but you could also look at the sign of the effect. Most people just report the magnitude, but it's often just as important to understand the direction of the effect.

  5. #5
    R must die
    Points: 25,023, Level: 95
    Level completed: 68%, Points required for next Level: 327
    noetsi's Avatar
    Posts
    4,536
    Thanks
    278
    Thanked 730 Times in 700 Posts

    Re: Negative effect size?

    Quote Originally Posted by trinker View Post
    Dason I disagree but maybe I don't understand.

    Cohen's d is an effect size and you can have a negative effect. If product B is worse than product A it will have a negative effect on the outcome of Y in comparison to A.
    I think you meant me as Dason did not post here (although its pretty complmentary of course to be mistaken for Dason)

    Right, as I noted it is negative (or positive) relative to the reference group. The effect of B on Y relative to A will generate a negative effect, the reverse would be true if you compare A on Y relative to B. It is an artificial number, the magnitude of the effect is the same. If you want to know how powerful an effect is (and thus how substantively important) there is no sign and indeed Cohen's D is usually discussed in terms of the absolute value.
    This was not what we did in logistic regression. Rather, we transformed the conditional expected value, and made that a linear function of X. This seems odd, because it is odd..

  6. #6
    Beep
    Points: 61,778, Level: 100
    Level completed: 0%, Points required for next Level: 0
    Awards:
    Discussion EnderPosting AwardCommunity AwardMaster TaggerFrequent Poster
    Dason's Avatar
    Location
    Ames, IA
    Posts
    11,114
    Thanks
    261
    Thanked 2,158 Times in 1,839 Posts

    Re: Negative effect size?

    Quote Originally Posted by trinker View Post
    Dason I disagree but maybe I don't understand.
    Ok. I'm sorry I said something. Oh wait I'm not cause I didn't.

    But I had the same objection to noetsi too.

  7. #7
    R must die
    Points: 25,023, Level: 95
    Level completed: 68%, Points required for next Level: 327
    noetsi's Avatar
    Posts
    4,536
    Thanks
    278
    Thanked 730 Times in 700 Posts

    Re: Negative effect size?

    *sob*

    I guess it depends on what is important. When I have seen effect size discussed they wanted to know how substantively important an effect size was which is tied entirely to its magnitude.
    This was not what we did in logistic regression. Rather, we transformed the conditional expected value, and made that a linear function of X. This seems odd, because it is odd..

  8. #8
    ggplot2orBust
    Points: 35,306, Level: 100
    Level completed: 0%, Points required for next Level: 0
    Awards:
    User with most referrers
    trinker's Avatar
    Location
    Buffalo, NY
    Posts
    3,949
    Thanks
    1,356
    Thanked 754 Times in 675 Posts

    Re: Negative effect size?

    Yeah I meant noetsi, I said dason. Sorry this whole bot thing has really got me on edge.
    "If you torture the data long enough it will eventually confess."
    -Ronald Harry Coase -

  9. #9
    TS Contributor
    Points: 5,883, Level: 49
    Level completed: 67%, Points required for next Level: 67
    jpkelley's Avatar
    Location
    Vancouver, BC, Canada
    Posts
    440
    Thanks
    17
    Thanked 90 Times in 84 Posts

    Re: Negative effect size?

    Quote Originally Posted by noetsi View Post
    *sob*
    I guess it depends on what is important. When I have seen effect size discussed they wanted to know how substantively important an effect size was which is tied entirely to its magnitude.
    Exactly right. Usually, I think authors have already shown graphs to indicate the direction, so it's only necessary to indicate the magnitude.

  10. #10
    R must die
    Points: 25,023, Level: 95
    Level completed: 68%, Points required for next Level: 327
    noetsi's Avatar
    Posts
    4,536
    Thanks
    278
    Thanked 730 Times in 700 Posts

    Re: Negative effect size?


    Quote Originally Posted by trinker View Post
    Yeah I meant noetsi, I said dason. Sorry this whole bot thing has really got me on edge.
    Skynet Dason....

    This was not what we did in logistic regression. Rather, we transformed the conditional expected value, and made that a linear function of X. This seems odd, because it is odd..

+ 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