STATS 6 CHOOSING PREDICTORS IN MR Flashcards
What is “collinearity”?
Two or more predictors are highly correlated and explain essentially the same variance in the outcome
Why is collinearity a problem?
(at least one predictor is redundant) Betas will be unstable, and their standard errors large, making it difficult to interpret how much different predictors impact the regression equation
What does the “tolerance” index represent?
Tolerance measures lack of collinearity
What is the simplest method to measure collinearity?
R-squared (describes the proportion of variance in X1 explained by other predictors)
How can R-squared be adapted to measure tolerance?
1 - R-squared = the tolerance value
Which tolerance values represent potential and serious problems?
<0.2 indicates potential problem
<0.1 indicates serious problem
Will SPSS take care of collinearity for you?
Hell no! SPSS only warns you if tolerance < 0.001.
You must request tolerance specifically!
What are possible solutions to collinearity problem??
Remove some predictors or combine highly correlated predictors together
What is a “suppressor” variable?
- improves overall prediction even though it is unrelated to outcome variable
- supresses irrelevant variance in other predictors and enhances their role in the regression model
What is the plane lesson example of suppressor variable?
If they had high spatial, numerical and mechanical scores… A low verbal score would mean higher ability to fly
How to spot the suppressor variable in regression?
Insignificant zero-order correlation, significant beta correlation
What is hierarchical regression?
Block entry
Creating a hierarchy of predictors
How do you compare two blocks in hierarchical regression?
Assess R-square change
What would we consider to be a good set of predictors in MR?
Important, theoretically relevant and based on good theory.
low correlations with each other and high correlations with the outcome