Chapter 11: Moderation, Mediation, and Multicategory Predictors Flashcards
grand mean centering
transformation of a variable by taking each score and subtracting the mean of all scores from it
moderation
occurs when the relationship between two variables changes as a function of a third variable
- two or more variables have a synergistic relationship
- effect of one predictor depends on the value / level of the moderator
- possible to include the combine effect in a statistical model
- statistical term for it is an interaction effect
moderator
variable that changes the size and/or direction of the relationship between two other variables
graphs: flat plane = no interaction; shifted/twisted plane = interaction
how do you create an interaction variable?
multiply the two predictors
if the regression coefficient term for the interaction is significant, there’s an interaction
once you detect an interaction effect, determine what the nature is since there are many
what are the three approaches to describing the nature of an interaction?
1) simple slopes analysis
2) Johnson-Neyman method
3) graph results of simple slope analysis
simple slopes analysis
describes the effect of the predictor on the criterion at three levels of the moderation (i.e., low, medium, and high on a trait; 1 SD below, 0 SD, 1 SD above)
johnson neyman method
similar to simple slopes
instead of decribing the effect of the predictor on the outcome at 3 levels of the moderator, the effect is described at many levels of the moderator
the output displays a “zone of significance” based on p
two ways to test for interactions in SPSS:
1) manually: transform variablels into deviations around a certain point. advantage is that we can more easily check for sources of bias
2) the process tool: it centers predictors for us, creates interaction term, produces simple slopes
mediation
situation when the relationship between a predictor variable and an outcome variable can be explained by their relationship to a third variable, called the mediator
partial mediation
predictor effects the outcome through a mediator but in addition the predictor also has a direct relationship (not through the mediator) with the outcome
full mediation
direct effect is not present, meaning the entire effect of the predictor on the outcome is through the mediator
statistical model for mediation
conceptual and statistical models match, unlike with moderation
1) Baron and Kenny (1986): historically populat, less direct method of testing for mediation, not ideal.
2) Sobel Test (1982): estimate the indirect effect and its significance, test for the sig of combined effects of a and b, works well in large samples
3) Preacher and Hayes (2004): use bootstrapping to create CI of the indirect effect (i.e., the effect of the predictor on the outcome that is mediated through the mediator)
- if the CI doesn’t include 0, that implies the effect isn’t 0 and that there is an effect
categorical predictors in regression
fairly straightforward (code 0 or 1)
we often want to include categorical predictors with more than two categories, so we need dummy coding
dummy coding steps
1) count the # of groups you want to recode and subtract 1
2) create as many new variables as the value you calculated in step 1
3) choose one of the groups as a baseline to compare all other groups to (control)
4) assign baseline group values of 0 for all dummy variables
5) for the 1st dummy variable, assign the value of 1 for the first group that you want to compare against the baseline group. assign all other groups 0 for this variable
6) for the 2nd dummy variable assign the value of 1 to the second group that you want to compare against the baseline. assign all other groups 0
7) repeat until you run out of dummy variables
8) place all dummy variables into the linear model in the same block
index of mediation
standardized measure of an indirect effect. in a mediation model, it is the indirect effect multiplied by the ratio of the SD of the predictor variable to the SD of the outcome variable
