+ Reply to Thread
Results 1 to 2 of 2

Thread: Single Negative Beta in Multiple Linear Regression

  1. #1
    Points: 248, Level: 4
    Level completed: 96%, Points required for next Level: 2

    Posts
    1
    Thanks
    0
    Thanked 0 Times in 0 Posts

    Single Negative Beta in Multiple Linear Regression




    Hi everyone

    I'm having trouble interpreting a single negative regression coefficient in a multiple linear regression. The situation is as follows:

    - 16 input variables >> 1 outcome variable
    - Individually, all 16 input variables are positively correlated with the outcome variables (all around .45)
    - In the multiple regression, 3 of the 16 input variables show significantly positive Betas, 12 seem to have no influence, and one is significantly negative
    - Tests for intercollinearity show no problematic values

    I don't know how to handle this one negative Beta: On the one hand, its culture variable is positively correlated with the motivation variable (.39). On the other hand, keeping the other 15 input variables constant, it has a negative effect on the outcome variable.

    To make a practical example: Let's say we measure how much students like Biology, Math, English, and Sports and compare it to their final overall grade. The correlation between liking a subject and the overall grade is positive for all subjects. But a multiple regression shows positive Betas for Math and English, and a negative Beta for Sports.

    How would you explain this?

  2. #2
    TS Contributor
    Points: 17,775, Level: 84
    Level completed: 85%, Points required for next Level: 75
    Karabiner's Avatar
    Location
    FC Schalke 04, Germany
    Posts
    2,541
    Thanks
    56
    Thanked 640 Times in 602 Posts

    Re: Single Negative Beta in Multiple Linear Regression


    In a study, it was found that higher income was related to higher death risk
    in the following year. After controlling for AGE (age was positively correlated
    with income, since e.g. people with more job ecperience etc. tended to earn
    more, and higher age also increased mortatlity risk, of course), income was
    negatively related to death risk (which wasn't surprising, low income is a
    well-know risk factor). So, after controlling for additional variables, the "true"
    effect of a certain predictor might emerge. Do a little research for "suppressor
    effect".

    Kind regards

    K.




    ), income was now

+ 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