Lecture 5: Moderation and mediation analysis

Question 1

What are moderators?

Answer 1

Research into who benefits the most from a therapy or under what conditions a therapy works best. It is a variable that affect the direction and strength between two other variables. So, it is a variable that influences the effect of therapy on treatment outcome

Question 2

What are mediators?

Answer 2

Research into the working mechanisms-> a variable that partly explains the relationship between two variables. Mediator is usually the first step in finding the mechanism of change by an intervention

Question 3

How would you represent a moderation in GLM?

Answer 3

y(i) = b0 + b1x1(i) + b2x2(i) + b3x1(i)x2(i) + error(i)

Question 4

How would would you conduct a continuous moderator in SPSS?

Answer 4

Multiple regression with the intervention, the moderator and the interaction term as predictors. Create a new variable for interaction in SPSS. If the interaction is significant, then there is a moderation effect

Question 5

How would you interpret the output?

Question 6

How to interpret B1?

Answer 6

difference in expected y if x1 increases by 1 while x2 is 0

Question 7

How to interpret B2?

Answer 7

difference in expected y if x2 increases by 1 while x1 is 0

Question 8

How to interpret b3?

Answer 8

difference in effect of x1 on expected y if x2 increases by 1

Question 9

How to interpret continuous moderation?

Answer 9

Look at the graph. How does the DV change for each group and the moderator?

Question 10

How to do a nominal moderation in SPSS?

Answer 10

Using a factorial ANOVA with intervention and moderator as between subjects-factors. If the interaction is significant then there is a moderation effect. Also if the regression lines are not parallel then the interaction is significant

Question 11

How do we interpret the relationship between mediators and the outcome?

Answer 11

Change in mediator must precede change in outcome but causal statements should not be made about mediators as there was no manipulation

Question 12

What is c?

Answer 12

It is the total effect so the relationship between the intervention and the dependent variable

Question 13

What is a*b?

Answer 13

The indirect effect so the relationship between the intervention and the DV occurring through the mediator

Question 14

What is c’?

Answer 14

The direct effect so the relationship between the intervention and the DV, without the mediator and should be less strong than with the mediator included

Question 15

How can mediation be tested?

Answer 15

Degree of mediation can be assessed by the estimate of an indirect effect by bootstrapping with PROCESS and the confidence interval. If the indirect effect is larger than 0 then there is mediation.

Question 16

How to interpret total effect output?

Answer 16

If value is negative, suggest that average DV values at post-test are lower after the intervention than after WLC. If positive, average DV values at post-test are higher after the intervention then after WLC

Question 17

How to interpret path b and the indirect effect?

Answer 17

Look under the outcome variable: DV.
If negative: the higher the mediator, the lower the DV if the IV remained constant.
If positive: the higher the mediator, the higher the DV if the IV remained constant
C’: the expected DV is lower after intervention than after WLC is the mediator remained constant

Question 18

How to interpret the indirect effect?

Answer 18

Remember to always look at the a and b values. First interpret the a value and then the b value like: EMDR increased sense of control at posttest compared to WLC (a) and an increased sense of control was associated with less expected PTSD at posttest (b)

Question 19

When is there partial mediation?

Answer 19

When the indirect effect is less than the total effect and when there is a direct effect

Question 20

How to report the indirect effect?

Answer 20

b = -2.78, 95% BCa CI [-5.38, -0.03]
If confidence interval (CI) contains 0, then no significant mediation

Question 21

What needs to be reported for each path?

Answer 21

1. the relationship between independent variable and mediator (path a): bvalue, tvalue, and p-value;
2. the total effect (path c): bvalue, tvalue, and pvalue;
3. relation between mediator and dependent variable (path b) and the direct effect (path c’): bvalue, tvalue, and pvalue;

Question 22

How to interpret a nominal moderator?

Answer 22

the factorial (2 Intervention x 2 Motivation) ANOVA shows a significant interaction effect between intervention and motivation, F(1, 50) = 4.92, p = .031. The beneficial effect of CT compared to WLC on change in panic symptoms between pre- and posttest was larger for highly motivated clients compared to low motivated clients.

Question 23

How to interpret a continuous moderator?

Answer 23

the treatment effect is moderated by external attribution at pretest. The
multiple regression analysis reveals a significant interaction effect between
intervention and external attribution, b = -1.95, t(47)= -2.87, p = .006. The effect of intervention (CT vs. WLC) on change in panic symptoms between pre- and posttest depends on the degree of external attribution at pretest; the lower external attribution at pretest, the larger the expected change in panic symptoms between pre- and posttest after CT compared to WLC

Question 24

How to interpret the indirect effect?

Answer 24

Step 4 (path a*b, indirect effect): Catastrophic cognitions at intermediate test significantly mediated the effect of intervention on change in panic symptoms b = 1.69, 95% BCa CI [0.33, 3.65]. Compared to WLC, CT reduced catastrophic cognitions at intermediate test (a) and less catastrophic cognitions at intermediate test were associated with a larger expected change (decrease) in panic symptoms (b).

Question 25

When do we test for a moderator and why?

Answer 25

Before in case the moderator can be influenced by the intervention (e.g., degree of severity). For other moderators, such as gender, age, etc., the measuring moment
does not matter.

Question 26

When would we use a mediator instead of a moderator?

Answer 26

whether the effect of intervention can be (partly) explained, in other words mediated, by catastrophic cognitions. Moreover, catastrophic cognitions are measured during intervention and may therefore be influenced by the
intervention. As a result, catastrophic cognitions can (potentially) explain the relationship between two variables (intervention and change in panic symptoms).

Question 27

How can we establish causality?

Answer 27

When the mediator precedes the DV and when the DV is measured at all time points with the mediator and when there is manipulation of the mediator.

Question 28

When would you use a moderator instead of a mediator?

Answer 28

A moderator is a variable that influences the relationship between two variables. Here, the question is whether the relationship between intervention (CT vs. WLC) and change in panic symptoms is influenced by the degree of external attribution. In addition, external attribution cannot have been influenced by the intervention (as the variable was measured at pretest) and therefore cannot explain (mediate) the relationship between intervention and change in panic symptoms.