L9 - moderation and mediation

Question 1

control variables

Answer 1

No control variables in a theoretical framework!

Question 2

Subsample analysis

Answer 2

Suitable when Z is clearly categorical, e.g.
(High vs medium tech)

1. For each subsample, regress Y on X and controls
   Statistical power is enhanced when subsamples have similar numbers of observations
2. Compare (verbally) the coefficient and statistical significance of X across subsamples

Question 3

continuous Z

Answer 3

Do not categorize!!!

Will reduce statistical power to detect moderating effect

Question 4

Moderated regression analysis

Answer 4

Suitable when Z is continuous

– For each different value of Z, there is a unique effect of X on Y.

Question 5

Possible moderating effects

Answer 5

X1 and X2 have independent effects on Y (a3=0) = no interaction
X1 and X2 have substitute effects on Y (a3<0) = negative interaction
X1 and X2 have complementary effects on Y (a3>0) = positive interaction

Question 6

Mean-centering

Answer 6

No statistic impact! centering on any other constant does not either.

However, log transformation is different!!

Question 7

Mediator vs moderator

Answer 7

Moderator > Z alters effect X on Y
Strengthen, weaken, change the sign of the X-Y relationship
Possible explanation for conflicting findings

Mediator = intervening variable > indirect effect of X on Y via M

Question 8

IV vs DV

Answer 8

IV = predictor  
DV = criterion

Question 9

Significance Baron & Kenny + critique

Answer 9

Baron & Kenny:
c’ insignificant = full mediation
c’ insignificant < absolute c = partial mediation

Critique:

- total effect should be significant - NO
- direct effect should be insignificant - NO
- Problems when multiple mediators
- Problems when correlated error terms > solution: simultaneous equations approach

Note:
Only requirement mediation = significant indirect effect

Question 10

GSEM vs OLS

Answer 10

GSEM = Generalized Structural Equation Model
Ø when Y and/or M are binary, multinomial, ordinal, or count

OLD
Y and M can be seen as continuous, and they are more or less normally distributed

Question 11

How test mediation

Answer 11

Panel data

Question 12

conditional process analysis

Answer 12

Moderated mediation = integrates mediation and moderation analysis

Question 13

Serial vs parallel mediation

Answer 13

Multiple mediation models can be parallel or serial