Please enable JavaScript.

Coggle requires JavaScript to display documents.

Regression (module 3&4) (Basics (Multiple regression (interaction…

- - - - not sig
        
        look at main effects
        
        Delete interaction
    - - Devide in high, medium, low according to SD
        
        B1 is centered
        
        Low
        
        substract SD from all scores
    - - or multiply p by amount of groups(?)
- - - - B
        
        difference between group means
        
        can change when adding more predictors
    - - VIF
        
        Should be <10
        
        otherwise centering
  - - - Multiply main effects
    - - look at colinearity #
        
        predictors have linear relationship with each other
        
        can increase SE
        
        loss of power
        
        Solution
        
        centering
        
        0, 1 becomes -1, 1
      - Main effects should still be included
- - - - VIF
        
        <10
    - - scatterplot
        
        Horizontal
        
        conspicuous values on predictor
        
        more problematic
        
        Vertical
        
        bad predictions
      - Studentized residuals
        
        outside -3 - 3
      - Centered leverage values
        
        ?
      - Cooks distance
        
        Should be <1
    - - robust against violations
      - do data transformation
        
        if not much changes it's ok
      - multimodality
        
        more problematic
        
        subgroups
    - - homogeneity of variance
      - scatterplot of standardized residual against standardized predicted values
        
        should be random
- - - - coefficients say how much groups differ from reference group
      - Reference group=intercept
  - - - Can also be used in regression by using -1, 0, 1 codes
        
        then B's represent group mean - grand mean
        
        Or other values that add up to 0
  - - - But only if they are dependent (correlated)?
    - - solutions
        
        Turn quantative variable int dichotomous
        
        median split
        
        high and low X
        
        twoway Anova
        
        if significant
        
        1 more item...
        
        Loss of data
        
        Look at simple slopes
        
        Effect of X at specific value of Z
        
        -1SD, mean(0 for centered), +1SD
        
        Needs to be calculated by hand
- - - - weakest
      - No number below 1
        
        change constant
    - - logX=Y
      - no numbers below 1
        
        constant
    - - strongest
      - Y^1= 1/Y
      - reverses ordinal order of scale values
      - Preparation
        
        multiply everything by -1
        
        add constant do make minimum 1
        
        take inverse of all values
- - - - per group
      - scatterplot
    - - check by adding relevant product terms
        
        multiply experimental factors
        
        Start with all interactions
        
        first look at second order
        
        remove if not sig
        
        Then look at others and remove least sig