Stats Question!

Discussion in 'Psychology [Psy.D. / Ph.D.]' started by parto123, Dec 17, 2008.

  1. parto123

    2+ Year Member

    Joined:
    Dec 25, 2006
    Messages:
    316
    Likes Received:
    0
    Status:
    Non-Student
    Hey stats people-

    im running a multiple regression with certain ocean personality characteristics as predictor variables and a parenting scale as the dependent variable.

    if i wanted to do either of the following, what is the best way to do so (on spss):

    1) control for gender and race demographics- (hierarchical regression?)
    2) see how these demographics moderate (or interact, or whatever the kids are calling it these days) the relationship between the personality and parenting variables.

    Thanks
     
  2. cmuhooligan

    10+ Year Member

    Joined:
    Jan 15, 2005
    Messages:
    312
    Likes Received:
    2
    Status:
    Psychologist
    1.) I would enter your control variables in the first step of a hierarchical regression, and then in the second step enter in your predictors.
    2.) To test for interactions (i.e., moderators) you would need to create interaction terms (e.g., genderXopenness). Then you would enter your main effects (e.g., gender, race, predictor variables) into the first step, and then put your 2-way interactions into the second step.
     
  3. blindblonde

    blindblonde U.S. citizen, Dutch Ph.D
    2+ Year Member

    Joined:
    Jan 21, 2007
    Messages:
    246
    Likes Received:
    0
    Status:
    Psychology Student
    1) I don't necessarily see the need for putting the control variables in Step 1. I think it would be important to see if the personality variable predicts the dependent variable at all in Step 1. You can then add the additional controlling variables to see the beta coefficients for the personality variables decrease with the other controlling variables. You can also look at the change in R-squared to see if the control variables add anything beyond the personality variables. I mean, either way works (this is a rather picky point). It just depends on how you want to look at the research question--do you want to know the effect of the predictor before you control for these variables or after you control for these other variables?

    2) Already addressed. :)
     
  4. cmuhooligan

    10+ Year Member

    Joined:
    Jan 15, 2005
    Messages:
    312
    Likes Received:
    2
    Status:
    Psychologist
    I don't think one needs to do this in step one. It is simply one approach that could be done.
     
  5. cmuhooligan

    10+ Year Member

    Joined:
    Jan 15, 2005
    Messages:
    312
    Likes Received:
    2
    Status:
    Psychologist
    I should also mention that if you are going to explore possible interactions, it is important that you center your interaction variables (to protect from possible multicolliniarity).
     
  6. blindblonde

    blindblonde U.S. citizen, Dutch Ph.D
    2+ Year Member

    Joined:
    Jan 21, 2007
    Messages:
    246
    Likes Received:
    0
    Status:
    Psychology Student
    True, I should have selected a better verb choice there. With stats in particular, there are multiple ways to tackle the situation at hand. :cool:
     
  7. parto123

    2+ Year Member

    Joined:
    Dec 25, 2006
    Messages:
    316
    Likes Received:
    0
    Status:
    Non-Student
    Thank you. This was helpful
     
  8. Jon Snow

    Jon Snow Senior Member
    10+ Year Member

    Joined:
    Aug 5, 2005
    Messages:
    3,225
    Likes Received:
    403
    The advantage the hierarchical gives you is seeing the model without your IVs of interest and then the impact of the IVs. I'd also consider calculating an f-change (an option if your using SPSS). Hierarchical also allows you to force the demographic variables in the first step and then the freedom of using stepwise or other methods in the second step. Consider running bivariate correlations first. If the demographic variables don't relate to your DV, then you don't need to control for them statistically.

    Moderator and mediator effects can be examined with partial correlations (this is calculated in the regression and should be on the output); but you can also run them separately (a partial correlation analysis).
     
  9. parto123

    2+ Year Member

    Joined:
    Dec 25, 2006
    Messages:
    316
    Likes Received:
    0
    Status:
    Non-Student
    Thanks for all the responses.

    A few more questions:

    1) To enter the ethnicity moderator into this regression, how would I code that. I have 4 groups (can this be done?)

    2) How about for Gender (0,1?)

    3) if I decide to go with performing just correlations instead of regressions, would I use a partial correlation for the gender/ethnicity moderation, or can I do interaction terms in correlation as well?

    Thanks
     

Share This Page