SMCR Kap 10-11

All None Speichern

• Confounder Variable that is not included in the model, but correlated with both IV & DV
• Reinforcer Adding this variable in the model makes the coefficient weaker/change sign
• Suppressor Adding this variable in the model makes the coefficient stronger
• Partial effect Unique effect of one variable, controlling for effects of other IV's
• Size of indirect correlation Product of correlation between predictor & 3rd, and between 3rd and dependent variable
• Use of regression Look at variation in scores on IV to predict variation of scores on the DV
• Controlling for a variable Adding it in the model
• Conceptual definition of controlling The regression removes the variation in the DV predicted by all other IVs, and then determines how well the remaining IV predicts the variation that is left
• Coefficient in regression model expresses Unique contribution of a variable on a prediction
• How to obtain partial effect Control for all other independent variables in our interpretation of a regression model
• Omitted variable bias Confounders in the model
• Relationship that confounder establishes Indirect correlation between predictor and dependent variable
• The larger the indirect correlation Increase of change in the regression coefficient if X is included as a new predictor
• Trick to avoid confounders Randomization in experiment conditions
• No systematic difference in group Experimental treatment not correlated with characteristics of the group
• Indirect correlation contradicts the effect of the predictor Effect is suppressed by the confounder
• Correlation has opposite sign of current effect It is a suppressing variable
• Regression tell us there is no effect of predictor We cannot rule out a suppressing effect
• Standardized reg.coe (in simple reg.model) equal to Correlation between predictor and outcome
• If a confounder is not added Effect gets mistakenly attributed to the effect of present predictors
• Spuriousness Part of effect due to confounding variable
• Reinforcers Establishes indirect correlation that has the same sign as the current effect of the predictor on the DV
• Causal criteria Non-spuriousness(commonly supported of cause and effect), time order, correlation
• Direct effect Supposedly causal effect of one variable on another
• Indirect effect Effect in which three or more variables affect each other in a causal order
• Total effect Sum of direct and indirect effects
• Mediator Variable that is both predictor, and predicted in the model
• Parallel mediation Model in which indirect include at most one mediator
• Serial mediation Model in which at least one indirect effect includes more than one mediator
• Curved arrow in regression model Indirect effect of X via Z on Y
• Standardized effect represents Predicted difference in SD's of Y, that results from a difference in Z
• LLCI Lower limit confidence interval
• ULCI Upper limit confidence interval
• Path model Second predictor mediates the effect of the first predictor on the outcome variable
• Size of indirect effect equals Product of direct effects that constitute the indirect effects
• Advantage of mediation Helps us think of various causal effects of a variable, on another variable
• Assumption Causal order, we cannot prove that the predictive effects are causal
• Criterion 1 of Mediation as a causal process Correlations/associations between variables
• Criterion 2 of Mediation as a causal process Time order between cause and consequence
• Criterion 3 of Mediation as a causal process Non-spuriousness, where an effect incorrectly includes the effect of a confounder(reinforcer)
• Causal diagram Contains names of variables with arrows pointing from cause to consequence
• Adding mediators Tool for getting more insight in the causal process
• Path model estimation Series of regression models
• Variable with at least one incoming arrow DV manifestation, estimate regression for each of them
• Unstandardized R tells us Sizes of direct effects in path model
• No inclusion of effects between variables No correlation/effects being 0 between them (thin arrows)
• Size obtainment of indirect effect Multiply direct effects (both unstandardized and standardized)
• Total unstandardized effect Sum direct and indirect effects
• Direct effect on outcome zero Effect of predictor wholly mediated
• Coming across full mediation Rarity, we usually detect partial mediations in our models
• No effect of mediators Direct effect is total effect
• Covariates in model Allowed to have an effect that can be caused by them
• Do not use standardized coefficients If predictor is dummy/dichotomous
• Every indirect effect can be wrong Effects in a path model can be confounded
• Causality or underlying construct? Mediators must be theoretically/conceptually different from predictor and outcome

All None Speichern