+ Reply to Thread
Results 1 to 7 of 7

Thread: Understanding pairwise correlation

  1. #1
    Points: 533, Level: 10
    Level completed: 66%, Points required for next Level: 17

    Posts
    50
    Thanks
    20
    Thanked 0 Times in 0 Posts

    Understanding pairwise correlation




    Maybe it is the current time, but I have some difficulties understanding what my tutor means (and he is gone on vacation so I can't ask).

    In my regression, I have six variables that reflect culture. Other studies used maybe two or three at most of these variables. Therefore he said: that these few variables can also reflect the other culture aspects that are not accounted for and that I can back this up with pairwise correlation & if these are positive run the regression only with one variable.

    so he wants me to a pairwise correlation between my six variables and if there is a positive correlation between two, run them separately in a regression.

    Other question: what if there is a negative correlation what does that mean then? why only positive correlation?

  2. #2
    TS Contributor
    Points: 14,811, Level: 78
    Level completed: 91%, Points required for next Level: 39
    Miner's Avatar
    Location
    Greater Milwaukee area
    Posts
    1,171
    Thanks
    34
    Thanked 405 Times in 363 Posts

    Re: Understanding pairwise correlation

    If you are referring to correlation between the explanatory variables, your tutor is probably referring to multicollinearity. This can occur anytime 2 or more of the predictor variables are correlated (positive OR negative). Testing the correlations prior to regression can help identify a POTENTIAL issue, but checking the Variance Inflation Factors (VIFs) in your regression model is more typically used. A VIF > 5 indicates that multicollinearity is present, and should be remedied. Dropping one of the correlated variables is one way of dealing with the issue.

  3. #3
    Points: 533, Level: 10
    Level completed: 66%, Points required for next Level: 17

    Posts
    50
    Thanks
    20
    Thanked 0 Times in 0 Posts

    Re: Understanding pairwise correlation

    Quote Originally Posted by Miner View Post
    If you are referring to correlation between the explanatory variables, your tutor is probably referring to multicollinearity. This can occur anytime 2 or more of the predictor variables are correlated (positive OR negative). Testing the correlations prior to regression can help identify a POTENTIAL issue, but checking the Variance Inflation Factors (VIFs) in your regression model is more typically used. A VIF > 5 indicates that multicollinearity is present, and should be remedied. Dropping one of the correlated variables is one way of dealing with the issue.
    The thing is that he told me this AFTER I run the regression with all predictor variables and reported the VIF (which are all between 1 and 2). He has me very confused on what he wants/his logic

  4. #4
    TS Contributor
    Points: 14,811, Level: 78
    Level completed: 91%, Points required for next Level: 39
    Miner's Avatar
    Location
    Greater Milwaukee area
    Posts
    1,171
    Thanks
    34
    Thanked 405 Times in 363 Posts

    Re: Understanding pairwise correlation

    If your VIFs are 1 - 2, your tutor is confused.

  5. #5
    Points: 533, Level: 10
    Level completed: 66%, Points required for next Level: 17

    Posts
    50
    Thanks
    20
    Thanked 0 Times in 0 Posts

    Re: Understanding pairwise correlation

    Quote Originally Posted by Miner View Post
    If your VIFs are 1 - 2, your tutor is confused.
    he did say that the VIF are all ok, before mentioning what I said in my first post...

  6. #6
    Points: 1,741, Level: 24
    Level completed: 41%, Points required for next Level: 59

    Posts
    230
    Thanks
    37
    Thanked 68 Times in 59 Posts

    Re: Understanding pairwise correlation

    Quote Originally Posted by Miner View Post
    If you are referring to correlation between the explanatory variables, your tutor is probably referring to multicollinearity. This can occur anytime 2 or more of the predictor variables are correlated (positive OR negative). Testing the correlations prior to regression can help identify a POTENTIAL issue, but checking the Variance Inflation Factors (VIFs) in your regression model is more typically used. A VIF > 5 indicates that multicollinearity is present, and should be remedied. Dropping one of the correlated variables is one way of dealing with the issue.
    I've also heard a VIF of around 10 is a red flag, but of course, multicollinearity issues can present at much lower VIFs. I would caution that "the correct remedy is to drop an independent variable from the regression." If your goal is prediction then MC is largely irrelevant as it has no negative impact on the model fit and prediction, therefore you don't need to drop a variable (and doing so might actually reduce the predictive ability of the model). However, if the goal is beta inferences and interpretations, then you do need to mitigate the effects of multicollinearity. Dropping independent variables until there is little evidence of an issue is one possible solution, but other methods, like centering, can sometimes remedy the issue.

  7. The Following 2 Users Say Thank You to ondansetron For This Useful Post:

    hlsmith (05-03-2017), Jazz3 (05-03-2017)

  8. #7
    TS Contributor
    Points: 14,811, Level: 78
    Level completed: 91%, Points required for next Level: 39
    Miner's Avatar
    Location
    Greater Milwaukee area
    Posts
    1,171
    Thanks
    34
    Thanked 405 Times in 363 Posts

    Re: Understanding pairwise correlation


    Quote Originally Posted by ondansetron View Post
    I would caution that "the correct remedy is to drop an independent variable from the regression."
    Quote Originally Posted by Miner View Post
    A VIF > 5 indicates that multicollinearity is present, and should be remedied. Dropping one of the correlated variables is one way of dealing with the issue.
    I did not say "the correct remedy is to drop an independent variable from the regression." I said dropping is one way of dealing with the issue.

+ 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