Multicolinearity Flashcards
VIF meaning
Since the SE is the square root of the variance, the SE is inflated by this square root. Concretely the VIF means that the SE has been doubled
This indicates that you fixed multicolinearity
SE went down
Types of multicollinearity
Artificial/ non essential
Natural / essential
Fixing non essential multicollinearity
Centering
Fixing essential multicolinearity
Drop one or more correlated predictors
Turn predictor into ratio (alcohol and welth, the more rich you are is proportionate to the expensive drinks you buy)
Composit variables, combines variables (sum / multiply correlated variables) instead GRV, GRAW just compute GREscore
Design a better experiment
Ways to dropp a predictor
Use stepwise to decide (not your best choice because you don’t think in advanced)
Eliminate causly dependent variables (age to predict education)
Hierarchical approach to eliminate causally dependent predictors
Drop theoretically unimportant predictors
Example of a composite variable
Height and weight…. BMI (takes your weight and divides that by your height so pum! 2 variables in one)
You get rid of a predictor that accounts for 0% of variance… what do you expect to happen to your R square and R adjusted?
R square: goes down because just adding increases it and just removing one decreases it
R adjusted: goes up because of penalty for having many predictors in the model
Multicolinearity changes______________ but doesn’t change_________
Increases the degree of uncertainty about your regression weights but doesn’t change your accuracy of predictability
PLS
Partial Least Squares. Uses a rotation of the prediction space for identifying predictors that are going to be better at predicting
Collinear is the wrong term.. the right term according to Mike is…
Confounded…“these things are confounded with one another”
Multicollinearity between continuous and categorical
VIF no longer valid. Must do Albelson’s law #7 have to move the couch to see the dust
