+ Reply to Thread
Results 1 to 2 of 2

Thread: (Urgent!) Overlapping concepts in regression analysis

  1. #1
    Points: 4, Level: 1
    Level completed: 7%, Points required for next Level: 46

    Posts
    1
    Thanks
    0
    Thanked 0 Times in 0 Posts

    (Urgent!) Overlapping concepts in regression analysis




    Hi all,

    I have an urgent question regarding my master thesis:

    I'm doing research on online political participation, which is measured by
    12 items with answer possibilities ranging from 'never' to 'sometimes' and 'often'.
    I've done a linear regression in SPSS, with 8 'classic' explanatory variables.
    6 of them are in a signifcant relation to online political participation.

    Many of these significant relations disappeared however when adding the variables 'online discussion' and 'offline discussion',
    measured by the question 'During the last year, how often have you:
    - discussed politics online
    - discussed politics offline'
    In both cases the options were: never - rarely - sometimes - often.

    I'm having serious issues interpreting this shift in significance. Does this mean that
    the variable 'online discussion' was always in the construct 'online participation',
    but now that it's substracted, the remaining items of online participation are no longer
    explained (or to a lesser extend) by the initial 'classic' 8 variables?

    I have another two methodological questions:

    1) Can someone please propose a good method to disentangle the concepts 'online participation' and 'online discussion', because there might be some overlap: some of these 12 questions on online participation were like 'have you commented on a news site/...blog/...forum' and another fourth was about sharing a political videoclip (you can imagine that doing so on for instance Facebook can lead to 'discussion' in the comment section below the video.

    2) This is answerable without digging into the subject, I figure:
    When after adding a variable that in substance partly overlaps with the dependent variable to an OLS regression, the relationships become no longer significant (eg. in the case of 'gender' significance dropped dramatically from 0.02 to 0.80), is such change genuine or not?


    Thank you in advance!

  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: (Urgent!) Overlapping concepts in regression analysis


    My guess is that you have signficant multicolinearity. That the two new variables are highly correlated with the other IV and with the DV and thus when you add them they eliminate the impact of the other IV. Essentially there is very little unique impact of the IV on the DV so they lost significance when the new variables were added. Whether that means the new variables you added are "part" of something is a theoretical not statistical question. All that statistics addresses is whether different variables covary.

    Can someone please propose a good method to disentangle the concepts 'online participation' and 'online discussion', because there might be some overlap:
    I think before you address that statistically you need to decide theoretically (or based on the existing literature) if they overlap. You are asking a theoretical question and trying to answer it with statistics - which can never happen.

    Are the definitions similar, are you measuring them the same way? If the items that measure one of these questions (variables) also load heavily on the other variable and (additionally) other sets of variables that are logicially not associated with online participation do not load heavily with online discussion then this suggest you are measuring the same factor with both. This is where convergent and divergent validity and structural equation models come into play.

    When after adding a variable that in substance partly overlaps with the dependent variable to an OLS regression, the relationships become no longer significant (eg. in the case of 'gender' significance dropped dramatically from 0.02 to 0.80), is such change genuine or not?
    I am not sure this is addressable by statistics - I don't even know what genuine means But you might look at the literature on moderating variables and spurious relationships to address this. Ultimately you have to make this decision based on what makes theoretical sense not correlation.

    To start with you should run a VIF test and see if you have high multicolinearity among your predictors.
    "Very few theories have been abandoned because they were found to be invalid on the basis of empirical evidence...." Spanos, 1995

+ 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