Please enable JavaScript.
Coggle requires JavaScript to display documents.
Ch 9: Mediation with regression analysis (criteria for causal relation…
Ch 9: Mediation with regression analysis
Partial effect
: controlling for all other ind variables in or regression model
how a multiple reg model controls the effect of an independent variable for the effects of all other independent variables
residual variation
variation in outcome scores predicted by only one variable
confounders
: variables that were not included in the model even though they are responsible for part of the effects
indirect correlation
: confounder establishes on indirect correlation between predictor and dependent variable
1st predictor can only predict proportion of 2nd predictor
1st IV can exactly predict 2nd IV (correlation=1)
positive direct correlation
: when none or both are negative
negative indirect correlation
: when 1 of the 2 correlations are negative and the other is positive
omitted variable bias
confounder added: IV changes reg coeff; size of change related to size of indirect correlation
when is partial effect
stronger
: indirect correlation has opposite sign of the direct effect
weaker
: when the indirect correlation has the same sign as the direct effect
2 types of confounders
simple: not controlling for there effect
partial: reg coeff if we controll
changes due to confounder (it is not added)
suppression
: indirect correlation contradicts the effect of the predictor (underestimated before confounder)
not included--> confounder negative effect is added to the predictors positive effect= SUPPRESSED (they were underestimated)
included--> suppression is eliminated
spuriousness
: adding a reinforcer variable to the regression model-->make entire predictor effect dissappear (over estimated before confounder)
reinforcer
: confounder added to model and all other effects are now weaker--> they were overestimated
adding mediations gives more insights about the causal process
full mediation
: when indirect effect via the mediator is qual to the correlation btwn x and y
criteria for causal relation
correlation (not causation )
time order
non-spurious correlation
path model
: includes both hypothesized (thick) and not hypothesized (thin) arrows