Week 4 - Moderation Flashcards
What is moderation?
Third variable relationship to test the indirect effect of IV and DV
X and Y relationship is observed to be different at different levels of the moderator (Relationship gone or weaker when Mod not present)
Moderation Product Term
Multiply the X and M variable to create a new variable
B-weight of this product term is the basis of moderation
Capture variance in Y not accounted for by the linear, additive effects of X and Mod. (Captures interaction effect/ joint effect between X and Mod)
Represent effect of moderation when other variables have been partialled out (unique contribution of the moderator)
How to show the from of moderation?
Simple effects
-B-weights
Plots
Assumptions of moderation?
Should not have a direct on Y
Should effect the relationship between X and Y
Has to be continuous Y
Can be any kind of X and M
Forms of moderation
Parallel = Relationship (interaction) does not exist Non-Parallel = Relationship (interaction) does exist (sig)
What are we testing for in moderation?
Does the product term (moderator) have additional variance over the individual effect of X and Y
What are 2 steps to test moderation?
1) Product term capture the assumed variance of the moderation
- Test using SE
- if more than 2 product term must be jointly tested
2) Probe the moderation (if significant) to establish the form
- Simple slopes and conditional effects (significant interaction mean lower order effect have to be qualified - simple)
When to run Factorial ANOVA or regression?
When Mod and X both categorical = ANOVA
When either Mod X, or both are continuous = MMR
Steps to MMR (5)
1) Derive/ establish product term
2) Set up regression model (Min X, Mod, Product term, control variables)
3) Evaluate the significance, size and strength of the product term (Use sr2)
4) Outline simple slopes that illustrate the form of the interaction in line with RQ and predictions (test using conditional effects test)
5) B-weights to find conditional effects (Simple slope allow plotting of moderation)
What is conditional Effect?
Slope of each predictor when the other is 0 (Slope X when Mod = 0 - Vice versa)
Change in Y for every 1 unit change in X or mod, when holding the other constant
What does product term add
How much X slope change as Mod changes
What does b1 gives us
Starting slope for X when X2 = 0
What does b3 give us
Tells us by how much the slope changes (Mod)\
Examining B1 and B3 give us the form of the moderation
Form moderation when IV and Mod are categorical
Normal plot - only one
Form moderation (regression lines) when IV continuous and Mod Categorical
Need to have separate plot for each level of the categorical predictor (separate plot for females and males) (Separate lines for different levels on the moderator)
Regression line for continuous calculated from regression equation (coeficients) (Fill in known fixed value of X and Mod to give points needed for the plot)
Form moderation for continuous Mod and IV
Need to select meaningful points to represent distinct classes of scores
Pair points on Continuous Mod with Continuous IV (to for regression line)
Hayes Pick a point approach
Set two points that will be meaningful fixed values on the Mod and IV
Loose convention 1SD above and below the mean
Derive predicted regression line from calculated scores from the regression equation with coefficients from the output
Statistical inference
M will be significantly different from 0 by implementing the pick a point approach
Can test whether X and Y slopes differ at each value on the Mod
Issues with moderation
Need to be guided by clear RQ and theory
- Knowing what variable is X and Mod
How and when variables measured in a study
Relationship may by synergistic (Two things work reinforce to have a better outcome) - not what moderation does/ no definite Mod
Give thought to model specification stage
Centreing
Subtract sample mean from each case before running MMR
Does not increase power or give more stable SE
Only change the b-weight for the constant (Can interpret the constant for that variable)
Does not affect anything else in the output
What is curvilinear analysis?
Looking for nonlinear additive effect
Increases not at a fixed rate
Testing for curvilinear effects
Square root of X as another possible predictor in the model
