5. Regression mediational and moderation Flashcards
What is the Baron and Kenny theory for?
Mediation
What do you write (in a path diagram) on the regression pathway?
the b values
What are the 4 steps to the Baron and Kenny model?
X on Y is sig. (SLR1)
X on M is sig. (SLR1)
M on Y is sig. (SMLR1)
But X on Y now not sig. (MLR1)
= full mediation
How do you test for significance of mediation after running the Baron and Kenny analysis?
Use a Sobel Test
What can you use to find the effect size of the mediation effect?
Haze Process Macros
What is a direct effect?
When X effects Y (SLR)
What is an indirect effect?
When X effects Y through M (MLR)
What is the equation for the mediation line?
M = intercept + a*X + e
a being the b1 value for X
What is the equation of the regression line where mediation is present?
Y = intercept + c’X + bM + e
where c’ is the b1 value for X on Y through M in a MLR and b is M on Y
Which b values do you use in a regression line equation?
The unstandardised b values
What are the 3 types of mediation?
Full
Partial
None
What does a dotted line in a path diagram mean?
Non sig. effect
When does partial mediation occur?
When the size of the p (sig. value) decreases but still remains sig. (so not full mediation)
When does no mediation occur?
When the path a or b are not sig. or close to 0
What is an exogenous variable?
Unrelated variables (e.g. no sig. effect… X predict Y and M predicts Y but X does not predict Y through M… e.g. no mediation)
When you have exogenous variables you have….?
No mediation
What is a spurious relationship?
Pearson r = sig.
But unstand. b value = 0
(no causal relationship)
What type of multiple regression analysis do you do in the third step of the barron and kenny process?
Simultaneous
What does the Adjusted R squared signify?
The percentage that accounts for the variance in the DV
What happens if the F value is sig. in an ANOVA?
The regression equation using the IV is sig. better than using the mean of the IV to predict the DV
What do F values compare in an ANOVA?
the mean and the model’s ability to predict the DV
What do you find in the coefficients table?
The unstandardised b values
What do the unstandardised b values tell us?
That when the IV increases by 1 point, the DV increased/ decreases by the b value*IV score
The b values can be either?
Sig. or non sig increase/ decrease
Given by the t-values
What are Cohen’s conventions for r?
.1
.3
.5
What are Cohen’s conventions for r sq?
1%
9%
25%
The adjusted R sq values (percentage of variance) can be either… which means?
Sig. or non sig.
Given by F values
(which means the model is either sig. or non sig. better than using the mean)
When put in a hierarchical multiple regression what can we see?
How each variable contributes to the variance and whether that contribution has a sig. effect
(whilst controlling for the other variables)
What is the formula for the sobel test?
a path * b path / SE a*b paths
What does the Haze Process Macro calculate?
The effect size (using the Confidence intervals using bootsrap)
Why is it good to use Haze process macros?
It gives more power in smaller samples
What is reported from Haze Macros output?
The Preacher and Kelly Kappa (K)
What is the maximum possible value of effect size (Kappa)?
1
How is Kappa intrepreted?
Cohen’s conventions or r sq
When is the indirect effect not genuine?
When confidence intervals of Kappa are -x and +x (include 0 point)
When does an interaction effect occur?
When X has an effect on Y when Z is present (better than adding both X and Z to get Y)
What is the interaction term also referred to?
The product term
or
The centred product term
Why do we need to centre the product term?
So that the scales are balanced
to standardise metrics (both carry same impact)
How do we centre a IV?
Standardise by:
IV - Mean of IV
What type of multiple linear regression do you use when using an interaction term?
Hierarchical (Blocking)
What term do you enter first into a hierarchical MLR when using an interaction term?
The normal IVS
(then in block to the Interaction Term)
IV1 and IV2
Then IV1*IV2
What can you conclude if after doing a MLR using an Interaction effect, the interaction term is non sig.?
Then the additive (main) effect is best (Normal linear regression will do e.g. DV = b0 +bIV1 + bIV2 + e)
What can you conclude if after doing a MLR using an Interaction effect, the interaction term is sig.?
Then predictors are synergistic (moderating each other)
What does synergistic mean?
Moderating (working together)
What are the df used when reporting F
Regression df Residual df (N - k)
What are the df when reporting t?
Residual df
N - k