Please enable JavaScript.

Coggle requires JavaScript to display documents.

Regression: Moderation A moderator must not be the causal result of the…

- - - - title column 1 as the predictor c (just an example, this is a centred/continuous variable)
        
        enter '=' into the cell below this title
        
        select the first cell from the predictor variable
        
        then type ' - AVERAGE' and select ALL of the cells in that predictor variable column
        the completed equation should be '= (cell) - AVERAGE'
        
        when this computed, drag the corner of the cell to create the averages for the rest of the data
      - title column 1 as the moderator c (just an example, this is a centred/continuous variable)
        
        enter '=' into the cell below this title
        
        select the first cell from the moderators
        
        then type ' - AVERAGE' and select ALL of the cells in that coumn with moderator variables (i.e. age)
        the completed equation should be '= (cell) - AVERAGE'
        
        when this computed, drag the corner of the cell to create the averages for the rest of the data
        
        NOW SAVE & return to JASP
  - - - 1. test for residuals/outliers
        
        go to 'statistics'
        
        look for the header 'residuals' and select:
        
        a) casewise disgnostics for outliers
        
        b) Keep the standard residuals as 3
        
        JASP will create a table called 'casewise diagnostic'
        
        IF there were any residuals, there would be a 'case number' of the residual
        
        1 more item...
      - 2. are residuals /outliers normally distributed?
        
        go to 'plots' dropbar
        
        select 'residual vs. histogram'
        
        ensure 'standardised residual' is still selected
        
        review it, does it look normal?
        
        NO
        
        1 more item...
        
        YES: Gaussian distribution (bell curve)
    - - 3. test for homoscedasticity (& look for linearity)
        
        under 'residual plots', select 'residual vs. predicted'
        
        a spotted graph should appear
        
        the dots should be randomly distributed (look like a potato), not a 'funnelled' triangular pattern
        
        if they are randomly distributed, this suggests there is linearity
        
        IF it is not linear, there will be an inverted U shape etc
      - 4. test for multicolinearity
        
        go to 'regression and click on 'correlation'
        
        only move the predictor (IV's) to the 'variables'
        
        Under the header 'sample correlation coefficient, select 'pearson's r'
        
        under 'additional ... select 'display pairwise'
        
        deselect 'report significance'
        
        1 more item...
    - - click on 'linear regression'
        
        place the outcome variable into 'deviation variable'
        
        mental note: 'dependent variable section' = outcome variable
        'covariates' = predictor variable
        
        place the centred variables that you did on excel (with the c beside them) in the 'covariates' column
        
        this will compute a 'coefficients' table
        
        go to 'statistics'
        
        select 'confidence intervals' as it indicates significiance and size of the relationship
      - review data
        
        look at the 'model summary' table
        what does the r2 / adjusted r 2suggest
        
        look at the ANOVA table, what does the p vaue suggest
        
        look at the interaction
        (this will tell us if we have a moderation)
        
        click on the 'model' dropbar
        
        from the 'components' header, select the predictor and moderator variables
        
        move them over to the 'model terms'
        
        this will add the interaction to the 'coefficients' table
        
        1 more item...
        
        breaking down interactions IF they are significant
        
        Go back to Excel
        
        create 2 columns titled .... high and ....low
        (depends on the moderator)
        
        (moderator title) low
        
        1 more item...
        
        (moderator title) high
        
        1 more item...
        
        we need to identify the low and high groups in the data
        
        Computing single slopes
        
        you may rename the linear regression to 'full model as 2 more will be made and it will be clearer
        
        go to 'regression'
        
        1 more item...
        
        go back to 'regression'
        
        1 more item...
        
        Go to the top bar and click on 'visual Modeling'
        
        IF this feature is not in the top bar, click the + sign (show modules menu)
        
        1 more item...