Multiple Regression Flashcards
how to do stepwise regression
- predictor that has the highest correlation with outcome gets entered first
- next strongest correlation gets entered next
- stop when remaining predictos no longer improve prediction of the outcome
why stepwise is a bad idea
- order that predictors are entered can depend on slight differences in predictive ability
- slight numerical differences can lead to major theoretical differences
adjusted R2
gives us some idea of how well the model generalizes
ANOVA test what
whether the model is significantly better at predicting the outcoming using the mean as a best guess
F-ratio
ratio of improvmenet in prediction that results from fitting the model relative to the inaccuracy that still exists in model
how are standardized b-values labels
beta1
what does standardized b value tell us
the importance of each predictor (the bigger absolute value = more important)
how to assess multicollinearity
VIF
standarized residula
residuals converted to z-scores
standard residular that is an outliner
> 3,29
standard residular that’s unacceptable error
> 2,58 (1% of cases)
standard residual that indicates poor model
> 1.96 (5% of cases)
influential cases
exert undue influence on the parameters of amodel
how do you know if a case has undue influence
Cook’s D > 1
Leverage
influnece of a predictor value will be stronger when it is further from the mena of that predictor
mahalanobi’s distance
distance of cases from the mean(s) of the prediction
covariance ratio
influence of a case on the variance of regression parameters
assumptions in multiple regression
additivity and linearity
predictors uncorrelated with external variables
outcome must be continous
non-zero variance
P-P plot
cumlative probability of standarized reisdual against cumulative probability of normal distribution
