Final Test Flashcards
Passing
What are mediation and moderation all about?
Relationships among three or more variables
What is mediation about?
The mechanism
The basic relationship
Between the predictor variable and outcome variable. Path c.
The a path
Between predictor variable and mediating variable.
The first regression: Unstandardized coefficients B and Std. Error
A value is beta
The b path
Two predictors (independent and mediating) on the dependent variable. Goal is B and SE for mediating variable on dependent.
The second regression: Unstandardized coefficients B and SE – makes sure it’s mediating row
: Standardized coefficients Beta weights for both predictors
B value is beta
Also collect number from zero order correlations
The direct effect
C prime. the beta between IV and DV when the MedV is included
What is a confidence interval?
An attempt to create a range that captures the amount of variability of that estimate. If the CI does not include the value of zero then it is statistically significant.
Standardised indirect effect
A times B. It is NOT a multiplication of the zero-order correlations because the second correlation does not reflect the presence of the other variable
The total effect mediation
the beta or raw correlation between the IV and the DV, c prime
Indirect to total ratio:
Indirect divided by total, answer is a percentage. Ratio scores are zero to 100. % of the basic relationship between the IV and the DV is explained by the indirect effect, unexplained is rest of ratio.
Multiple mediator model
Cannot be done with simple regression but done with structural equation modelling. Amos loaded with SPSS
Variable types in mediation
Typically done with three continuous variables, can also do when IV is dichotomous.
Can do mediation when either or both the MedV and DV is dichotomous, but this requires logistic regression, and it’s complicated.
Wald’s Test
The Wald’s test is like the F-test or t-test: The Exp(B), also known as the odds ratio, represents the probability that this particular outcome will occur.
Key point: The B and se are on a different scale than in linear regression
Longitudinal mediation
C prime path between IV1 and DV3
A path between IV1 and MV2
B path between MV2 and DV3
What is moderation about?
Moderation looks at associations of variables within groups (WHO), do different groups have different relationships between the IV and DV
Similarities between mediation and moderation
At the basic level, both involve 3 variables
Both can be done with ordinary regression in SPSS
Both examine the impact of a third variable on a 2-variable relationship
Differences between mediation and moderation
Moderation features an interaction (two Ivs are multiplied) term
Moderation is not a causal model
IV and ModV do not need to be correlated in moderation
Graphing is important in moderation
How do we do a moderation?
An interaction term is added to the basic model (on top of the 2 main effects) - regression with three predictors. So we have Iv, ModV and IV*ModV all leading to DV
Constraints on regression based moderation
IV: Must be continuous
ModV: May be continuous or categorical
DV: Must be continuous
Dummy coding basics
Dummy coding - must have a zero code.
You will have one less dummy variable than the number of categories.
You need to designate one category as the ‘comparison’ group
(all zeros)
Be sure that the ‘absence’ code is zero.
Interaction term in moderation
Multiply IV by ModV to create the new variable
Placement of variables when graphing moderation
IV is across on x axis
DV is up on y axis
ModV: Lines depict the slopes of the groups, slopes indicate the direction and strength of the association
Interpreting simple slopes!
Look at simple slope values and p values
What did Aiken and West have to say?
They argued that centring the IV and ModV before multiplying them to create the interaction term was advantageous. It has now become ‘standard practice’.
Centring is subtracting the mean of a variable from all scores on that variable. Yields a mean of zero and the same SD.
What did Kromrey and Foster-Johnson have to say?
They showed that centring has no impact on the significance of the interaction term nor on the graphed pattern of the moderation result
Moderation: Buffer
the moderator fosters lower levels of the negative DV. Highest level of moderator is flattest slope.
Moderation: Exacerbator
the moderator fosters higher levels of the negative DV. Lowest level of moderator is flattest slope.
Moderation: Enhancer
the moderator fosters higher levels of the positive DV. Lowest level of moderator is flattest slope.
Moderation: Dampener
the moderator fosters lower levels of the positive DV. Highest level of moderator is flattest slope.
Things to avoid doing in moderation
A failure to properly dummy-code categorical ModVs;
Confusion about what types of variables are proper ModVs;
Inaccurate graphing of the interaction result (in whatever graphing facility that you use);
Unfamiliarity with simple slope analyses;
Inexperience with making interpretations of moderation results; and
An inclination to use causal language in interpretation.
What is a quadratic polynomial?
A quadratic polynomial is the same thing as the IV moderating itself. In essence, it tells us that the slope of the association between the IV and the DV changes depending on the value of the IV, i.e., it’s not a constant slope.
Types of moderation
Linear moderation: X and Y and Z
A Quadratic: X2 and Y
Quadratic Moderation: X and Y and Z2
Moderation of a quadratic: X2 and Y and Z