interpretation of residual plot in multiple regression

Home Forums Default Forum interpretation of residual plot in multiple regression

Viewing 5 posts - 1 through 5 (of 5 total)
  • Author
    Posts
  • #3021
    ssamdani
    Member

    Hi all,

     

    I am trying to test the assumptions of multiple regression. I am currently looking at the properties of my residual plots. am slightly confused by this one. Can I say by looking at the attached plot that there is a linear relationship (probably with a negative slope) between my residuals and predicted values? and is it reasonable enough to say that there isn’t much hetroscadasticity.

    #3025
    Nino Rode
    Participant

    What I see is that you have six levels of dependent variable and that analysis (maybe exactly becuse you don’t have an interval variable) doesn’t predict much… What is R squared for this analysis?

     

    What you have got in this plot is that low responses on your dependent variable have negative resids and high have positive resids. In other words, you didn’t explain much with your analysis. For heteroscedascity I would rather look at residuals by values (not predicted) of dependent variable.

    #3024
    ssamdani
    Member

    Thanks for the reply Nino. The R square for this analysis was 42% (not very far from the usual trend in the type of work I am doing). It was a hierarchical regression in spss that was done in 5 levels.

    do you think the relationship is somewhat linear? also do you think it’s possible for a DV and IVs to share a linear relationship as well as a quadratic relationship at the same time? any thoughts?

    S

     

    #3023
    Nino Rode
    Participant

    You obviously have steps in your DV and continuous prediction: of course the relationship between the residuals and predictions is linear. That doesn’t mean much.

    I’m not sure what you mean with “sharing” a relationship. There sure can be a linear and quadratic relationship between the IV and DV. To look for it just fit something like:

    DV = IV + IVsquared

    or

    DVsquared = IV

    or

    DVsqared = IV + IVsquared

    However it would be good to standardise your vars to avoid too big numbers at squared values…

     

    #3022
    ssamdani
    Member

    thanks Nino :). I think I get what you mean. this was useful

Viewing 5 posts - 1 through 5 (of 5 total)
  • The forum ‘Default Forum’ is closed to new topics and replies.