Lecture 15 - Mediation & moderation Flashcards
(?) Describe the considerations on the complicated reality
General:
- Relationship between variables are often an oversimplification of real relationship: Not always direct, sometimes mediator or moderator
- Remember to mention how much of effect that is direct
- Only researcher make right choice. Program dont know mediation/moderation
Contingency factors:
- Factors changing how direct relationship works
Boundary conditions:
- If effect occur “unless” something. Shift relationship
- Reference to moderating
Interfering & transmitting factors:
- Factors crucial for transporting effects
- Reference to mediating
(!) Describe mediation & problems to it
General:
- IV indirectly affect DV through another variable: Mediator
- Interviening variable
- Partial or full mediation: full = only significant x to y effect if absent
- Correlation-theoretical reasoning is crucial
- Eg. Income enable taste preference for eco milk
Problems:
- Mediation effect w. opposite sign can balance direct effect out: Insignificant total effect
- Common methods variance bias: Especially for cross-sectional data
- Sobel test unreliable: Always report bootstrapped results
- Direct effect not always addressed by theory
- Dont combine cross sectional data & mediation: Causality
(!) Describe the conceptual model of mediation
Testing:
Old school:
- 1. Direct effect of IV on DV
- 2. Direct effect of IV on mediator: Set one IV as DV
- 3. Enter all at same time
Sobel test:
- Test significance of indirect effect
- Works well for big samples: z-test
- Determine if mediator carry the effect of IV to DV
Bootstrapping
- Even more novel
- Compare w. parametric result: Assumption such as normal dist.
__________
Evaluation:
- Partially standardized effects
- Completely standardized indirect effect: Comparable with other standardized effects
(!) Describe moderating & problems to it
General:
- Interactions between IV´s affecting DV: Moderator
- Easy if dummy: Work as switch
- If continuous variable: Use +/- 1 std
- Keep IV in model when moderator included
- No common method variance bias
- Direct effect of coefficient no special meaning if moderator incl.: Show effect on other IV when another IV is 0
- Interested in IV effect on different moderator levels
- Eg. Both sustainability concern & income affect purchase of eco milk
Problems:
- Measurement error of interaction is higher
- Require balanced sample w. substantial group size at all levels
- Sample must include full range of possible values: Problematic if non-random
- Artificial dishotomization: Loss of info & nature if two mutual excluding groups
- Remember why & when to use mean centering: Interpretation
- Using average effect if crossover interaction may balance effect out
(!) Describe the conceptual model of moderation
General:
- Test if significant interraction between IV & moderator
What centering changes:
- Centering ensure no multicollinearity: Easier interpretation
- Centering ensure interaction coefficient is unchanged
- Base coefficient = Effect of IV when sample is average
- Centering = At mean level
- Dont center dummy variables
Simple slope analysis:
- DV values predicted by IV at different moderator levels
- Bootstrap std. error: Allow nonparametric test of significance: Normality irrelevant
- Pick level to calculate IV-DV relationship
- Same as splitting sample in two separate regressions: Moderator either 0 or 1
- Read interaction effect in combination with base effects
- Graph the effect: Steeper line = more effect
(!) Give examples on mediation & moderation
Example 1:
IV: Participative budgeting
DV: Performance
Mediation:
- Participative budgeting –> Budget commitment –> Performance
Moderation:
- Budget participation + personality type –> Commitment to budget goals
__________
Example 2:
IV: Autonomy: Dansk. Selvstyre
DV: Performance
Positive moderation: Task complexity
Cross-over moderation: Self control. Since half positive, half negative
Mediation: Autonomy –> Scheduling efficiency –> Performance
Chain mediation: Autonomy –> Responsibility –> Motivation –> Performance
(?) Describe other causal forms than mediator & moderator