Lecture 5 Flashcards
Regression II
Video 1
Why do you use multiple regression?
Because it can control for confounding variables. It can examine the joint effect of multiple predictors
Video 1
What is a cofound/ covariable?
A variable that influences the DV and IV. You can decide the cofound in advance and collect the appropriate data
Video 1
What is the result of adding a covariate?
The SE will lower
Video 1
What does the formula for mutliple regression look like?
y = a + biX1i + b2X2i + b3*X3i + Ei
Video 1
What is the multiple correlation coefficient?
R, the correlation observed between DV and the prediction model.
Video 2
How do you evaluate multiple regression?
With R^2 (how well the variables can be explained), multiple t tests, when the F test does not equal the t test or when there is parsimony
Video 2
What is parsimony?
When a simpler model is possible you use the most simple model
Video 2
Does the beta coefficient equal correlation?
No, it does not
Video 2
What do you do in a nested regression model?
It is comparable to ANOVA, which you can test by eg only doing b2=0
Video 2
What is the F-change?
The F test is compared to a simpler and a more complex model (SPSS click option, enter/next)
Video 2
What is the R^2 adjusted?
When there is an adjustment made for the number of predictors (better comparable to models)
Video 2
When are coefficients changged?
When something is added/removed from the model.
Video 2
What are assumptions made in multiple regression?
The same as in simple regression, with the addition of no (perfect) multicollinearity: check the correlation between IVs using the VIF
Video 2
What is VIF?
The Variance Inflation Factor
Video 2
What formula is used to calculate VIF?
1/(1-R^2)
Video 2
What do the values of 1, 5 and 10 mean in relation to VIF?
VIF = 1: there is no correlation
VIF > 5: very correlated, consider removing (R^2>0.8)
VIF > 10: remove (R^2>0.9)
Video 2
Does each predictor have its own VIF?
Yes
Video 3
What does the predictor depend on?
On the group and whether there is an interaction
Video 3
What is a focal variable?
When there is only on IV, this is placed on the y axis in a graph
Video 3
What is the price of choosing a complex model?
There is a lower df, it can be overfit (too many factors are playing a role, which makes it more difficult to interpret)
Video 3
What is the consequence of using more tests?
It is more likely that there is significance by chance
Video 3
What is the chance of observing a type I error rate per test?
Increases with 0.5 per test. Calculate by 1-.95^n
Video 3
When do you use the Bonferroni correction?
When you have to correct for the multiple testing. You divide the aplha by the number of tests
Notes
What does unique covariance mean?
That X1 and X2 both independently explain y
